{"meta":{"query_hash":"a1e852829d09","filters":{"venue":"Software Quality Journal"},"cohort_total":51,"direct_labels_cover":0,"predictions_cover":51,"exported":51,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/a1e852829d09","api":"https://metacan.xera.ac/api/v1/cohort?venue=Software+Quality+Journal"},"results":[{"id":"W1537009764","doi":"10.1023/a:1021634422504","title":"Specification-based Testing for Gui-based Applications","year":2002,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Windsor","funders":"","keywords":"Computer science; Programming language; Java; Formal specification; Test harness; Finite-state machine; Graphical user interface testing; Test case; Implementation; Software engineering; Keyword-driven testing; Graphical user interface; Software; User interface; Software development","score_opus":0.15669253087361484,"score_gpt":0.34311759476107645,"score_spread":0.1864250638874616,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1537009764","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08443998,0.0002099128,0.90920925,0.00022665148,0.000027081382,0.000105036066,0.00006280753,0.003925744,0.0017935943],"genre_scores_gemma":[0.7707315,0.00016002901,0.22663471,0.00012998212,0.000016458605,0.000121906654,0.00039025812,0.0005181055,0.0012970802],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9917468,0.003893953,0.0005416868,0.00047186489,0.0028192815,0.0005263248],"domain_scores_gemma":[0.96368086,0.025857164,0.0014210842,0.004634395,0.0039638756,0.00044256973],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0054929513,0.0008168603,0.0006877073,0.0012634395,0.0005354482,0.0015794631,0.002686498,0.001520533,0.002552784],"category_scores_gemma":[0.027141431,0.00060969766,0.0009058754,0.0008836767,0.0011878381,0.002482311,0.0012744301,0.0016151698,0.00054338184],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023094683,0.001657608,0.021104006,0.00091909256,0.00022895145,0.0015771616,0.0015261712,0.20680846,0.11059142,0.0969352,0.0068290676,0.54951346],"study_design_scores_gemma":[0.00017314433,0.000539932,0.0017806689,0.00007252803,0.00009616488,0.0004973882,0.00013023261,0.9248952,0.048129056,0.021856459,0.0017926267,0.000036635938],"about_ca_topic_score_codex":0.0038927067,"about_ca_topic_score_gemma":0.0040530595,"teacher_disagreement_score":0.0054929513,"about_ca_system_score_codex":0.00074179604,"about_ca_system_score_gemma":0.0013866661,"threshold_uncertainty_score":0.029049873},"labels":[],"label_agreement":null},{"id":"W1537911176","doi":"10.1023/a:1025869312943","title":"Usability Meanings and Interpretations in ISO Standards","year":2003,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Usability and User Interface Design","field":"Computer Science","cited_by":454,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University; École de Technologie Supérieure","funders":"","keywords":"Usability; Computer science; Usability engineering; Software engineering; System usability scale; Component-based usability testing; Usability goals; Software quality; Standardization; Usability inspection; Software; Software development; Human–computer interaction; Programming language","score_opus":0.0318959053260117,"score_gpt":0.3304311199025112,"score_spread":0.29853521457649945,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1537911176","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10096501,0.012782606,0.460182,0.055185825,0.0043058125,0.0003448981,0.0004862175,0.0009966261,0.36475107],"genre_scores_gemma":[0.9077974,0.0027150363,0.07425328,0.0024970379,0.0011425951,0.0005547377,0.00034087148,0.0007038075,0.009995207],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.93015796,0.04330248,0.0073169316,0.0024353964,0.01518766,0.0015995073],"domain_scores_gemma":[0.8962694,0.07120963,0.004875939,0.008717247,0.017803801,0.001124061],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.029404044,0.0012811774,0.0010885223,0.012303336,0.0053374604,0.019320194,0.0020606206,0.004944146,0.0039746137],"category_scores_gemma":[0.100231074,0.0019443026,0.0012013911,0.008429482,0.034086086,0.029960942,0.0048567974,0.008429641,0.0009348049],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015628702,0.000012721966,0.00030683688,0.000110126166,0.000006756978,0.00007801827,0.02869817,0.00010745687,0.00024287478,0.96066505,0.0016790851,0.008077345],"study_design_scores_gemma":[0.000026134134,0.000038170925,0.0009534327,0.00083610765,0.000045665278,0.00025820092,0.017027723,0.0014283811,0.0006989219,0.934264,0.044364203,0.000059129765],"about_ca_topic_score_codex":0.0035326954,"about_ca_topic_score_gemma":0.0026750928,"teacher_disagreement_score":0.029404044,"about_ca_system_score_codex":0.005849048,"about_ca_system_score_gemma":0.005072492,"threshold_uncertainty_score":0.1555053},"labels":[],"label_agreement":null},{"id":"W1571880163","doi":"10.1023/a:1023764510838","title":"Towards a Software Change Classification System: A Rough Set Approach","year":2003,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Rough Sets and Fuzzy Logic","field":"Computer Science","cited_by":19,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada; Manitoba Hydro","keywords":"Computer science; Rough set; Data mining; Software; Machine learning; Artificial intelligence; Benchmarking; Software deployment; Software engineering","score_opus":0.16672560978481335,"score_gpt":0.3304675520946508,"score_spread":0.16374194230983743,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1571880163","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009103654,0.00035974584,0.98824614,0.0005471456,0.00004344363,0.00015373928,0.000121962854,0.0004925697,0.0009316714],"genre_scores_gemma":[0.11920177,0.0003504312,0.8791389,0.00017013383,0.0000729996,0.0002137596,0.00033866352,0.000045767847,0.00046759614],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9936167,0.0020710097,0.00082256855,0.0006856248,0.0025830523,0.00022101961],"domain_scores_gemma":[0.99120516,0.0038127867,0.0008910267,0.00081523997,0.0030337018,0.00024206635],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007878764,0.00094543124,0.0029291972,0.007026715,0.001450885,0.0065837596,0.0030747398,0.002017551,0.0011185543],"category_scores_gemma":[0.01981588,0.00073644606,0.0023783238,0.0036612134,0.0013385076,0.004033832,0.0025335166,0.0028530292,0.0008500119],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031814267,0.00046767914,0.009792278,0.0008197553,0.0007477626,0.00027778314,0.00147149,0.11761416,0.009038283,0.098155044,0.00892933,0.75236833],"study_design_scores_gemma":[0.000088285655,0.00030700088,0.0048489417,0.00023982971,0.00043843416,0.00032573877,0.0007161297,0.81977665,0.004344435,0.16201055,0.006743574,0.00016044415],"about_ca_topic_score_codex":0.0040678577,"about_ca_topic_score_gemma":0.0031287635,"teacher_disagreement_score":0.007878764,"about_ca_system_score_codex":0.0017865782,"about_ca_system_score_gemma":0.002296324,"threshold_uncertainty_score":0.0416674},"labels":[],"label_agreement":null},{"id":"W1967828228","doi":"10.1007/s11219-009-9082-y","title":"Improving design-pattern identification: a new approach and an exploratory study","year":2009,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":41,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Polytechnique Montréal","funders":"","keywords":"Identification (biology); Computer science; Constraint (computer-aided design); False positive paradox; Data mining; Resource (disambiguation); Resource constraints; Machine learning; Engineering; Distributed computing","score_opus":0.09743104232199579,"score_gpt":0.355633719165051,"score_spread":0.2582026768430552,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1967828228","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.183565,0.00027690033,0.80658805,0.00064250216,0.000053971504,0.0013791277,0.00023396757,0.0013056075,0.005954901],"genre_scores_gemma":[0.2767835,0.0002198894,0.7186904,0.0002578679,0.000027218948,0.000776261,0.0003394391,0.00017068397,0.0027348085],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98791444,0.005494983,0.00090152473,0.001944293,0.003476036,0.0002687038],"domain_scores_gemma":[0.9596115,0.024232216,0.002116673,0.0066988855,0.006864079,0.00047668096],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00658013,0.0014080937,0.0013000619,0.0037499212,0.0014604185,0.002963346,0.0024263149,0.0018273193,0.003942651],"category_scores_gemma":[0.038381644,0.0006715159,0.0012138746,0.0024333019,0.0014702387,0.007280332,0.0038932594,0.001860721,0.0010576682],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005452438,0.0039534904,0.037552677,0.0016336852,0.00021513959,0.0005376416,0.013606517,0.003788567,0.045769766,0.0103226,0.0024773742,0.8795973],"study_design_scores_gemma":[0.001729308,0.010616765,0.108052894,0.0014170781,0.002280945,0.010636152,0.049176484,0.4302506,0.22320312,0.0910166,0.071033,0.0005871531],"about_ca_topic_score_codex":0.001407343,"about_ca_topic_score_gemma":0.0030826603,"teacher_disagreement_score":0.00658013,"about_ca_system_score_codex":0.0010933132,"about_ca_system_score_gemma":0.002187964,"threshold_uncertainty_score":0.034799457},"labels":[],"label_agreement":null},{"id":"W1968155288","doi":"10.1007/s11219-014-9250-6","title":"Empirical analysis of factors affecting confirmation bias levels of software engineers","year":2014,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":31,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Confirmation bias; Software; Identification (biology); Metric (unit); Computer science; Software bug; Software engineering; Risk analysis (engineering); Engineering; Psychology; Social psychology; Operations management; Programming language; Business","score_opus":0.11496069501911305,"score_gpt":0.3726311614194243,"score_spread":0.25767046640031127,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1968155288","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.998722,0.000058582795,0.00032451237,0.000091068825,0.000003708844,0.000015267246,0.000025301684,0.000008773422,0.00075075135],"genre_scores_gemma":[0.9995227,0.00002107618,0.000258308,0.000018323148,0.000003348886,0.0000063207317,0.000025865666,0.0000027893816,0.000141261],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98340607,0.007709969,0.0021393604,0.0009850311,0.0046378328,0.0011217976],"domain_scores_gemma":[0.43437976,0.42543274,0.084752604,0.011521673,0.033094537,0.010818721],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.018796057,0.00028120118,0.00026249597,0.0029167095,0.00089913496,0.0018840859,0.000820833,0.0011232675,0.0032243165],"category_scores_gemma":[0.22558497,0.0003173257,0.0003516003,0.0014934579,0.0011171615,0.0011754888,0.00093506096,0.0011140597,0.00042321876],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003126813,0.00021995681,0.988655,0.000041963405,0.000043132677,0.00009293914,0.0022557445,0.00010042047,0.00089330936,0.00018879904,0.00010219727,0.0070938403],"study_design_scores_gemma":[0.000030031348,0.0005137597,0.99143124,0.000041275034,0.000070015674,0.000301821,0.0045406786,0.00073609897,0.0015615647,0.0003034323,0.00044444765,0.000025580683],"about_ca_topic_score_codex":0.002632102,"about_ca_topic_score_gemma":0.003292751,"teacher_disagreement_score":0.9812039,"about_ca_system_score_codex":0.0010985704,"about_ca_system_score_gemma":0.0035202256,"threshold_uncertainty_score":0.099404216},"labels":[],"label_agreement":null},{"id":"W1968405738","doi":"10.1007/s11219-006-9009-9","title":"A method to elicit architecturally sensitive usability requirements: its integration into a software development process","year":2007,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Usability and User Interface Design","field":"Computer Science","cited_by":20,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Usability; Usability goals; Usability inspection; Usability engineering; Computer science; Software engineering; Cognitive walkthrough; Usability lab; Web usability; Heuristic evaluation; Component-based usability testing; Pluralistic walkthrough; Human–computer interaction","score_opus":0.06656961381106619,"score_gpt":0.39863685788634606,"score_spread":0.3320672440752799,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1968405738","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016283307,0.000069513175,0.9787387,0.00019367476,0.000029100363,0.0008318388,0.000093092916,0.0010753591,0.0026853662],"genre_scores_gemma":[0.06672862,0.0000766942,0.92954487,0.00007550704,0.000007677231,0.001258249,0.00010154597,0.00017076818,0.0020360595],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99140304,0.0039731464,0.00050702924,0.0008376318,0.0031468836,0.00013234414],"domain_scores_gemma":[0.9704503,0.020263007,0.00148513,0.0036503284,0.003777515,0.0003736683],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005853315,0.00084147527,0.00071808044,0.0022124567,0.00080931,0.001702721,0.0009930421,0.0015015547,0.0028562406],"category_scores_gemma":[0.0230413,0.0007786208,0.000617238,0.0012690919,0.001049804,0.0022144427,0.002050522,0.0013566349,0.0012546846],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032588272,0.0014227891,0.0068139397,0.0013707522,0.00007813707,0.00029009164,0.009533457,0.0023675158,0.13168544,0.028954094,0.0050621186,0.8120959],"study_design_scores_gemma":[0.0013016168,0.0046323575,0.047056496,0.0021582274,0.00082482357,0.007843718,0.01255706,0.3526062,0.31046644,0.10534499,0.154162,0.0010460729],"about_ca_topic_score_codex":0.0007758604,"about_ca_topic_score_gemma":0.0018770732,"teacher_disagreement_score":0.005853315,"about_ca_system_score_codex":0.00063017407,"about_ca_system_score_gemma":0.0022963632,"threshold_uncertainty_score":0.030955613},"labels":[],"label_agreement":null},{"id":"W1971146998","doi":"10.1007/s11219-015-9271-9","title":"On the use of design defect examples to detect model refactoring opportunities","year":2015,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":39,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure","funders":"","keywords":"Code refactoring; Computer science; Software bug; Set (abstract data type); Context (archaeology); Similarity (geometry); Data mining; Software; Machine learning; Reliability engineering; Artificial intelligence; Engineering; Programming language","score_opus":0.5757960133181265,"score_gpt":0.376091517068576,"score_spread":0.19970449624955044,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1971146998","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.706314,0.0010182303,0.27146843,0.0012117048,0.000102684884,0.0003657509,0.0008565382,0.005059336,0.013603247],"genre_scores_gemma":[0.8073189,0.00028933867,0.18964007,0.00017610003,0.000020242784,0.00004294398,0.00061800535,0.00025896088,0.0016355434],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99556476,0.0013715901,0.00029585842,0.000572738,0.0019613125,0.00023364653],"domain_scores_gemma":[0.8982179,0.08204725,0.0039110514,0.0070357914,0.007878882,0.0009091658],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0033243657,0.0006584266,0.00058569515,0.004504628,0.0006319375,0.0017216933,0.0016189297,0.0021355879,0.0016625315],"category_scores_gemma":[0.0509396,0.00047051595,0.0005198001,0.0013559117,0.00068413746,0.0023741752,0.0014346903,0.0010448517,0.00053822284],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021972863,0.0014121564,0.16460659,0.0006672396,0.00032796408,0.002227332,0.001885288,0.042310324,0.05122132,0.009788552,0.0069095665,0.71644646],"study_design_scores_gemma":[0.00015290211,0.00060714415,0.043815624,0.00027602026,0.00032077043,0.0025585026,0.0007563421,0.8865381,0.04831105,0.010284001,0.0062337257,0.00014589312],"about_ca_topic_score_codex":0.0059739584,"about_ca_topic_score_gemma":0.01222961,"teacher_disagreement_score":0.0059739584,"about_ca_system_score_codex":0.0004486651,"about_ca_system_score_gemma":0.00081034366,"threshold_uncertainty_score":0.017581105},"labels":[],"label_agreement":null},{"id":"W1979334159","doi":"10.1007/s11219-007-9024-5","title":"The applicability analysis of ISO/IEC 9126 series of standards in telecom sector: the example of softswitch category of software","year":2007,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure","funders":"","keywords":"Softswitch; Software; Quality (philosophy); Computer science; Grid; Code (set theory); Telecommunications; Engineering; Software engineering; Operating system; Telephony; Mathematics","score_opus":0.030915266403321102,"score_gpt":0.3234580157601795,"score_spread":0.2925427493568584,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979334159","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6697305,0.001518371,0.21635473,0.0017451069,0.000108098015,0.0004357578,0.0007992071,0.00081367936,0.10849451],"genre_scores_gemma":[0.96714157,0.00043062324,0.0275619,0.00009522326,0.000029237368,0.00008157058,0.0005651504,0.0000717765,0.00402284],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9935022,0.0013313971,0.00037311064,0.0004096934,0.004031942,0.0003515941],"domain_scores_gemma":[0.98517704,0.0046684993,0.0010239154,0.0013771568,0.0075878394,0.00016562486],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0049557346,0.0003281732,0.000310758,0.005642984,0.0010821223,0.0018629787,0.0008768439,0.001019293,0.0024578024],"category_scores_gemma":[0.0133201275,0.00014027451,0.00066691707,0.004188194,0.0010594653,0.0020408507,0.00083468534,0.0008148177,0.000570834],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00054206693,0.00047778245,0.08095854,0.00066738296,0.00010024582,0.0017361359,0.004098958,0.038054843,0.053136542,0.47215608,0.008561087,0.33951026],"study_design_scores_gemma":[0.000047850615,0.0010185838,0.17743708,0.00063738093,0.00032913068,0.0021020025,0.007215315,0.37462586,0.07667575,0.28494453,0.074808985,0.000157586],"about_ca_topic_score_codex":0.009030515,"about_ca_topic_score_gemma":0.0061447658,"teacher_disagreement_score":0.009030515,"about_ca_system_score_codex":0.0020952218,"about_ca_system_score_gemma":0.002692092,"threshold_uncertainty_score":0.026208758},"labels":[],"label_agreement":null},{"id":"W1986024814","doi":"10.1007/s11219-013-9203-5","title":"BlackHorse: creating smart test cases from brittle recorded tests","year":2013,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Blackberry (Canada); Western University","funders":"","keywords":"Computer science; Keyword-driven testing; Java; Test case; Test Management Approach; Manual testing; Code coverage; Regression testing; Test (biology); Software engineering; Graphical user interface testing; Reliability engineering; Test harness; Programming language; Software; Engineering; Software development; User interface; Software construction; Machine learning","score_opus":0.044124431323995406,"score_gpt":0.31495280582406704,"score_spread":0.27082837450007163,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1986024814","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.033777513,0.00021508275,0.82454026,0.00018116015,0.00011557808,0.0004656759,0.0010966307,0.13537996,0.004228118],"genre_scores_gemma":[0.2942387,0.00024183569,0.6738008,0.00021027063,0.000045758028,0.0005203633,0.0046327976,0.020613696,0.005695749],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9977412,0.0005917949,0.00015543713,0.00036742017,0.00096922595,0.00017488837],"domain_scores_gemma":[0.98835886,0.007225937,0.00072148436,0.0022320584,0.0011453927,0.0003162838],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0023490363,0.0024561945,0.0009229177,0.0027395228,0.00049844844,0.0019686327,0.0033642473,0.0015302007,0.013573176],"category_scores_gemma":[0.018810036,0.001498685,0.0014864603,0.0010803312,0.0011842557,0.0029031676,0.00281084,0.0016184564,0.0032854243],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014502716,0.0006839823,0.015740883,0.0015009085,0.00043128702,0.0048092166,0.0023514875,0.10821217,0.060018845,0.018552974,0.0529854,0.7332625],"study_design_scores_gemma":[0.00037221352,0.000397229,0.0033766734,0.0002009579,0.00016384978,0.0016141195,0.00034071432,0.86888295,0.07875921,0.021585992,0.02416231,0.00014371693],"about_ca_topic_score_codex":0.0022102871,"about_ca_topic_score_gemma":0.0038655482,"teacher_disagreement_score":0.013573176,"about_ca_system_score_codex":0.0005201297,"about_ca_system_score_gemma":0.00093828625,"threshold_uncertainty_score":0.04540676},"labels":[],"label_agreement":null},{"id":"W2001599472","doi":"10.1007/s11219-010-9128-1","title":"An industrial case study of classifier ensembles for locating software defects","year":2011,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":67,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"Sabancı Üniversitesi","keywords":"Computer science; Software; Classifier (UML); False alarm; Software quality; Data mining; Domain (mathematical analysis); Ensemble learning; Embedded software; Machine learning; Artificial intelligence; Reliability engineering; Software development; Engineering; Operating system","score_opus":0.2236840213549557,"score_gpt":0.3779922822970453,"score_spread":0.1543082609420896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2001599472","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92701125,0.0003644177,0.0681788,0.0003270322,0.00004784302,0.00018857248,0.0001990621,0.00054325425,0.003139879],"genre_scores_gemma":[0.9472494,0.000100067016,0.05086465,0.000035583893,0.00001198632,0.000035296976,0.00015688197,0.000045595472,0.0015004908],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9974268,0.0009018117,0.00016063856,0.00039168616,0.00092753547,0.00019146343],"domain_scores_gemma":[0.98154956,0.011882583,0.0006998042,0.0025172767,0.0029522656,0.00039849692],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0038478682,0.0006649975,0.0006426304,0.0015652886,0.0010413994,0.0011298647,0.0019106513,0.0020956686,0.0015853302],"category_scores_gemma":[0.013429443,0.00031128168,0.0005537263,0.0015837582,0.0006263921,0.0010696582,0.0007609113,0.00087570254,0.00050237385],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023078131,0.0038892815,0.111221366,0.00083137787,0.0003215936,0.009728532,0.002760554,0.21614277,0.0465234,0.004251115,0.0075352346,0.59448695],"study_design_scores_gemma":[0.0003210168,0.0040647783,0.033370852,0.00010182652,0.00029792258,0.006298017,0.0022498749,0.87591225,0.06423763,0.0044330982,0.008612101,0.00010066007],"about_ca_topic_score_codex":0.003948894,"about_ca_topic_score_gemma":0.006650547,"teacher_disagreement_score":0.003948894,"about_ca_system_score_codex":0.0007788204,"about_ca_system_score_gemma":0.00059896696,"threshold_uncertainty_score":0.020349681},"labels":[],"label_agreement":null},{"id":"W2001932992","doi":"10.1007/s11219-010-9127-2","title":"Assessing the maintainability of software product line feature models using structural metrics","year":2010,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Advanced Software Engineering Methodologies","field":"Computer Science","cited_by":132,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Athabasca University; National Research Council Canada","funders":"","keywords":"Maintainability; Software product line; Feature model; Software metric; Computer science; Software; Data mining; Software quality; Feature (linguistics); Reliability engineering; Set (abstract data type); Software engineering; Software development; Engineering","score_opus":0.11787304923974923,"score_gpt":0.4112353565501069,"score_spread":0.2933623073103577,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2001932992","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9494597,0.000068522924,0.04956371,0.000044567205,0.000005346129,0.000025563471,0.00011335163,0.0002116811,0.00050767284],"genre_scores_gemma":[0.9868902,0.000024562942,0.012740985,0.0000030686565,0.00000314277,0.000011759392,0.00022379118,0.00001719727,0.00008528392],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9969836,0.0010063455,0.00024802552,0.000271766,0.0013787871,0.000111502224],"domain_scores_gemma":[0.9694603,0.019033734,0.0043755625,0.003062344,0.0037509927,0.0003170923],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0052664746,0.0005260176,0.000354077,0.0029293227,0.00030308464,0.0011902761,0.00078691234,0.00068851124,0.00047194437],"category_scores_gemma":[0.036886342,0.0003397757,0.0007357453,0.0013705688,0.0003302947,0.002294556,0.0006191192,0.00048082814,0.00009625007],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00068263034,0.00054231455,0.32949054,0.000175142,0.00059199653,0.0001282419,0.00050642085,0.43125597,0.03117234,0.0030902347,0.0005034279,0.20186082],"study_design_scores_gemma":[0.000022257203,0.00085445656,0.054469075,0.00001636108,0.00014330335,0.00007958245,0.00010733774,0.93393373,0.008235273,0.0018554763,0.00025823893,0.000024965211],"about_ca_topic_score_codex":0.0052725086,"about_ca_topic_score_gemma":0.005806211,"teacher_disagreement_score":0.0052725086,"about_ca_system_score_codex":0.0009958341,"about_ca_system_score_gemma":0.000711841,"threshold_uncertainty_score":0.027852118},"labels":[],"label_agreement":null},{"id":"W2015215117","doi":"10.1007/s11219-006-9218-2","title":"On evaluating the layout of UML diagrams for program comprehension","year":2006,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":30,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Class diagram; Sequence diagram; Applications of UML; Story-driven modeling; UML tool; Computer science; Unified Modeling Language; Communication diagram; Programming language; Activity diagram; Software engineering; Flowchart; Readability; Class (philosophy); Program comprehension; Use Case Diagram; Object Constraint Language; Software; Software system; Artificial intelligence","score_opus":0.08799721980182837,"score_gpt":0.4182070965023177,"score_spread":0.3302098767004893,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2015215117","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6277957,0.0009175699,0.35761485,0.00040044263,0.00008516542,0.00044033947,0.00043100625,0.005814361,0.0065006036],"genre_scores_gemma":[0.7412964,0.0004133077,0.25390756,0.000080454134,0.00003669525,0.00014572807,0.0013574797,0.0007477857,0.0020145334],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99279803,0.004285528,0.0004497195,0.00064354716,0.0015801619,0.0002429421],"domain_scores_gemma":[0.8286266,0.15125868,0.0044054016,0.004115639,0.010062542,0.0015312122],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005629472,0.0012273496,0.00091106293,0.0035655198,0.00076099136,0.002717255,0.0012300086,0.0017397495,0.005346701],"category_scores_gemma":[0.10537454,0.0005318857,0.00061865203,0.001971996,0.00070100266,0.0051402305,0.0015345797,0.0008492525,0.0008472698],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025289415,0.0012624698,0.039576214,0.0010913343,0.00016311473,0.00023177818,0.004312283,0.044927735,0.07201586,0.0050852653,0.004218918,0.82458615],"study_design_scores_gemma":[0.0005493962,0.0044887797,0.065997966,0.00035116833,0.00046132438,0.00040598752,0.0047840076,0.82631606,0.0774894,0.012839931,0.006148449,0.0001675745],"about_ca_topic_score_codex":0.007655989,"about_ca_topic_score_gemma":0.011606581,"teacher_disagreement_score":0.007655989,"about_ca_system_score_codex":0.0012532478,"about_ca_system_score_gemma":0.0015966386,"threshold_uncertainty_score":0.029771805},"labels":[],"label_agreement":null},{"id":"W2015762629","doi":"10.1023/b:sqjo.0000013356.15791.3d","title":"Requirements for Maintaining Web Access for Hearing-Impaired Individuals","year":2004,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Digital Accessibility for Disabilities","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Hearing impaired; Computer science; Dream; Sight; Visually impaired; Interface (matter); User interface; Human–computer interaction; World Wide Web; Multimedia; Audiology; Operating system; Psychology; Medicine","score_opus":0.2112760787145806,"score_gpt":0.4694726119693898,"score_spread":0.2581965332548092,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2015762629","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8844189,0.00046544897,0.043660842,0.00467143,0.00020275441,0.001373301,0.0009238047,0.0022256116,0.062057916],"genre_scores_gemma":[0.95217496,0.00022102863,0.040384922,0.00044087696,0.0001002296,0.0004093396,0.0009584199,0.00030926138,0.005000974],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9879118,0.0021740575,0.0021659434,0.00033906955,0.0062744003,0.0011346459],"domain_scores_gemma":[0.88556045,0.0644988,0.009510883,0.005524645,0.02959836,0.0053068316],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00968174,0.00032399138,0.00043491126,0.0016216278,0.0017715616,0.0035172293,0.0016271198,0.0018229835,0.0059987595],"category_scores_gemma":[0.08193964,0.0005106255,0.00050867285,0.00071284804,0.0005963365,0.0025832425,0.0019327641,0.0010341238,0.0016215761],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002063689,0.0022065996,0.2542095,0.0018671359,0.00014026876,0.007539452,0.028651088,0.013663479,0.1611277,0.02833065,0.046433218,0.45376733],"study_design_scores_gemma":[0.00028016476,0.003998412,0.66175014,0.0017826386,0.0005647229,0.02226676,0.035038088,0.07033208,0.07187544,0.015347588,0.116277486,0.00048642716],"about_ca_topic_score_codex":0.00823018,"about_ca_topic_score_gemma":0.008702663,"teacher_disagreement_score":0.00968174,"about_ca_system_score_codex":0.0010631947,"about_ca_system_score_gemma":0.0028150172,"threshold_uncertainty_score":0.051202595},"labels":[],"label_agreement":null},{"id":"W2021780242","doi":"10.1007/s11219-011-9153-8","title":"AoURN-based modeling and analysis of software product lines","year":2011,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Advanced Software Engineering Methodologies","field":"Computer Science","cited_by":37,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Software product line; Feature model; Software engineering; Domain engineering; Domain (mathematical analysis); Computer science; Domain analysis; Feature-oriented domain analysis; Feature (linguistics); Notation; Product (mathematics); Systems engineering; Software; Model-driven architecture; Model transformation; Software development; Engineering; Programming language; Artificial intelligence; Software construction","score_opus":0.1482975576947659,"score_gpt":0.3486422211534545,"score_spread":0.2003446634586886,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021780242","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04759533,0.00021765132,0.9474449,0.00013403637,0.000017949797,0.000041347477,0.0001431965,0.0006207296,0.0037848055],"genre_scores_gemma":[0.79852694,0.0004667969,0.19302337,0.000035710284,0.000022381731,0.00012431966,0.00040952256,0.00021664599,0.007174312],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9993216,0.00022400895,0.00003753371,0.000102010665,0.00024612428,0.00006878966],"domain_scores_gemma":[0.9987318,0.00063787657,0.00019176697,0.00014864221,0.00025365438,0.000036198093],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010996625,0.00054109463,0.00046758942,0.0013368445,0.0005330398,0.001783409,0.0016846715,0.0009086287,0.0023686737],"category_scores_gemma":[0.0031288124,0.0005172441,0.00094765873,0.0010999972,0.0006182919,0.0015776407,0.00081182184,0.00085292093,0.0005681688],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000026841653,0.000040625233,0.001287081,0.000038495822,0.000018733615,0.000054846496,0.00010050394,0.94088984,0.001885326,0.03769443,0.00021987465,0.017743317],"study_design_scores_gemma":[0.000001536566,0.000005217849,0.0001035499,0.0000024589265,0.0000037082254,0.0000054809507,0.0000050191193,0.9951226,0.00033166257,0.0039855894,0.00043120433,0.000001987668],"about_ca_topic_score_codex":0.018975345,"about_ca_topic_score_gemma":0.01358479,"teacher_disagreement_score":0.018975345,"about_ca_system_score_codex":0.0014280064,"about_ca_system_score_gemma":0.0012971566,"threshold_uncertainty_score":0.0377298},"labels":[],"label_agreement":null},{"id":"W2022628966","doi":"10.1007/s11219-006-9010-3","title":"Measuring size, complexity, and coupling of hypergraph abstractions of software: An information-theory approach","year":2007,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":34,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"University of Waterloo; Mississippi State University; National Science Foundation","keywords":"Computer science; Software metric; Software sizing; Theoretical computer science; Hypergraph; Software; Data mining; Software development; Software quality; Software construction; Programming language; Mathematics","score_opus":0.07791100068699204,"score_gpt":0.32054010421443907,"score_spread":0.24262910352744704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2022628966","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6275755,0.00031038464,0.36911482,0.00030108308,0.000013731207,0.00009003819,0.00018747164,0.00032652568,0.0020804028],"genre_scores_gemma":[0.95623696,0.0000880609,0.04324974,0.00001765961,0.000019343897,0.000047075137,0.00014906643,0.000034001503,0.00015802437],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9948606,0.0017192672,0.0003437964,0.00068384653,0.0021278297,0.0002645543],"domain_scores_gemma":[0.87652314,0.097189665,0.010913453,0.008965368,0.0046238545,0.0017845267],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0050661764,0.00071063376,0.0011799363,0.0084818825,0.00087642757,0.0032324907,0.0017661217,0.0016091027,0.00082130724],"category_scores_gemma":[0.062482525,0.0008340247,0.0012097256,0.004522404,0.0027013996,0.012154467,0.002491046,0.0015565953,0.000093587005],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012468698,0.0009309625,0.19986613,0.0007465841,0.0010651087,0.00030790854,0.0039941045,0.38445655,0.03541561,0.14655481,0.0010567665,0.22435863],"study_design_scores_gemma":[0.00004581951,0.0003265275,0.043197405,0.000042159532,0.000298363,0.00016807651,0.0006671237,0.79965013,0.0111824535,0.1438069,0.00050298916,0.0001120323],"about_ca_topic_score_codex":0.0027625312,"about_ca_topic_score_gemma":0.0022770728,"teacher_disagreement_score":0.0084818825,"about_ca_system_score_codex":0.0018766131,"about_ca_system_score_gemma":0.0010297376,"threshold_uncertainty_score":0.026792824},"labels":[],"label_agreement":null},{"id":"W2025855180","doi":"10.1007/s11219-009-9088-5","title":"An organizational maturity model of software product line engineering","year":2009,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":37,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Software product line; Capability Maturity Model; Software Engineering Process Group; Service Integration Maturity Model; Product management; Product engineering; Software quality management; Software development; Engineering; Process management; Software engineering; Systems engineering; Software development process; Product (mathematics); New product development; Computer science; Knowledge management; Software; Product design; Business; Marketing","score_opus":0.029049896642323363,"score_gpt":0.3048473787423666,"score_spread":0.27579748210004323,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2025855180","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.117724404,0.000984656,0.8186139,0.0052026957,0.00008129523,0.0003530405,0.00021205944,0.00046017018,0.056367803],"genre_scores_gemma":[0.85978276,0.0005728417,0.13556162,0.00012199784,0.00003497497,0.000223107,0.00019724981,0.000045368364,0.0034600531],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9976829,0.00094240997,0.00017154166,0.00018952889,0.0007572206,0.0002563096],"domain_scores_gemma":[0.9935992,0.0025788138,0.00076259143,0.00040593406,0.002257836,0.00039569574],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0037985207,0.0005401938,0.00031074884,0.002567029,0.0008248181,0.0038596569,0.0012991372,0.0017141014,0.0021331497],"category_scores_gemma":[0.011727352,0.00042129427,0.000612773,0.0018780294,0.00091983756,0.0073571634,0.0013580607,0.0018529623,0.00080406794],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008918017,0.0003287986,0.015613361,0.00016987619,0.000074761665,0.00019208086,0.002164419,0.05559445,0.0027080474,0.7913611,0.0025934281,0.1291106],"study_design_scores_gemma":[0.000056739114,0.00031550255,0.014078193,0.00026896183,0.0001234626,0.00020469978,0.0015082565,0.58512425,0.0013204191,0.38229638,0.01462649,0.00007663079],"about_ca_topic_score_codex":0.006977576,"about_ca_topic_score_gemma":0.00636824,"teacher_disagreement_score":0.006977576,"about_ca_system_score_codex":0.0029764627,"about_ca_system_score_gemma":0.003274825,"threshold_uncertainty_score":0.021595836},"labels":[],"label_agreement":null},{"id":"W2029160271","doi":"10.1007/s11219-014-9232-8","title":"Toward a mature industrial practice of software test automation","year":2014,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Automation; Software engineering; Test (biology); Software; Computer science; Software testing; Engineering; Operating system","score_opus":0.06356689489556629,"score_gpt":0.33149529574143893,"score_spread":0.26792840084587266,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2029160271","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016066948,0.0067835376,0.87433636,0.044877943,0.00079804234,0.00042027293,0.00006197855,0.0021802746,0.05447453],"genre_scores_gemma":[0.24724823,0.0061793304,0.72984844,0.0052776616,0.0011327298,0.00040129636,0.00017547031,0.00061259966,0.009124179],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.96194947,0.01554881,0.0026602645,0.0034492172,0.015254747,0.0011374387],"domain_scores_gemma":[0.8314699,0.058136884,0.0061206794,0.035534594,0.058013044,0.010724879],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05554742,0.0010386048,0.0010135232,0.0050752433,0.0030134604,0.012932111,0.0059090266,0.0068437327,0.0045121973],"category_scores_gemma":[0.08901803,0.00096243486,0.0008303528,0.003026246,0.012941385,0.013966969,0.0069312397,0.014195956,0.002497121],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000090862086,0.0010479428,0.005745389,0.0007070448,0.00006568574,0.00020404917,0.0038117187,0.0060147946,0.005351363,0.57903713,0.018872973,0.379051],"study_design_scores_gemma":[0.00015609058,0.00082336954,0.009265673,0.002939301,0.000064076245,0.0014083036,0.0036128066,0.03206409,0.010811883,0.58198726,0.35667974,0.00018739213],"about_ca_topic_score_codex":0.0023456463,"about_ca_topic_score_gemma":0.0023549867,"teacher_disagreement_score":0.05554742,"about_ca_system_score_codex":0.0043500704,"about_ca_system_score_gemma":0.015996315,"threshold_uncertainty_score":0.29376632},"labels":[],"label_agreement":null},{"id":"W2029277954","doi":"10.1007/s11219-014-9230-x","title":"Predicting defective modules in different test phases","year":2014,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Benchmark (surveying); Reliability engineering; Computer science; Predictive modelling; Software; Data mining; Machine learning; Engineering; Programming language","score_opus":0.028114770166436563,"score_gpt":0.31264121370854725,"score_spread":0.2845264435421107,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2029277954","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9915222,0.00013489599,0.006933991,0.000035607754,0.000012444676,0.000017356364,0.0004883252,0.0005694918,0.00028583308],"genre_scores_gemma":[0.99416757,0.000031839016,0.0041408436,0.000014827473,0.000004826716,0.0000067858255,0.0011646856,0.000047503647,0.00042112538],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9987613,0.00020438927,0.00013299522,0.00026915857,0.00042694694,0.0002052227],"domain_scores_gemma":[0.9723652,0.015611279,0.004287153,0.0015386605,0.004799532,0.0013981981],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016448434,0.0009835205,0.00056924776,0.0044500143,0.00025899053,0.00088676723,0.0010891935,0.0012291074,0.0016603012],"category_scores_gemma":[0.01782574,0.0003915818,0.0010376514,0.0014293704,0.00036525086,0.00093338545,0.0004956094,0.00063769065,0.00065198395],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00116196,0.00049726065,0.8908943,0.00009911325,0.000165032,0.00034238107,0.000105703104,0.030315684,0.01134613,0.00020236734,0.0007930036,0.06407714],"study_design_scores_gemma":[0.00006016117,0.0018466464,0.5972244,0.0000493361,0.0003767913,0.001106547,0.00028226717,0.37306926,0.024079992,0.0011519899,0.0006991134,0.000053389427],"about_ca_topic_score_codex":0.005593194,"about_ca_topic_score_gemma":0.0069248034,"teacher_disagreement_score":0.005593194,"about_ca_system_score_codex":0.0005506662,"about_ca_system_score_gemma":0.0005966031,"threshold_uncertainty_score":0.011121273},"labels":[],"label_agreement":null},{"id":"W2044659433","doi":"10.1023/b:sqjo.0000013359.71560.47","title":"Multi-Project Management in Software Engineering Using Simulation Modelling","year":2004,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":28,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Software project management; Computer science; Project management; Interdependence; Systems engineering; Project management triangle; Software; Software development; Engineering; Software construction","score_opus":0.11692953807586136,"score_gpt":0.3729647007733453,"score_spread":0.25603516269748394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2044659433","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07252593,0.00044305448,0.91398096,0.000423134,0.000055135417,0.0001476287,0.00006968042,0.00040093914,0.011953406],"genre_scores_gemma":[0.8726203,0.0005312504,0.122765176,0.000042679574,0.000024483914,0.00029008798,0.000111597365,0.00008339998,0.0035310374],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99671423,0.0022729132,0.00016257438,0.00020933118,0.0005009353,0.0001400935],"domain_scores_gemma":[0.99248075,0.005691545,0.0005072559,0.00046925052,0.0006194846,0.00023174312],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0033491056,0.0006707189,0.0010260381,0.0013640488,0.0008744494,0.0024342372,0.0014421255,0.0017558126,0.0031722095],"category_scores_gemma":[0.008502817,0.0008087068,0.001136464,0.0016006388,0.0008800153,0.0028145781,0.0014094039,0.0012762892,0.00043489077],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000038601647,0.000073969386,0.0012264987,0.00004113464,0.0000337832,0.000046796275,0.00013988452,0.9633601,0.00047505705,0.02356501,0.00016689475,0.010832271],"study_design_scores_gemma":[0.0000094100915,0.000024596142,0.00017643276,0.000014138795,0.000010029714,0.000012831698,0.00002929498,0.9910924,0.00026547827,0.0076169763,0.00073974865,0.00000864134],"about_ca_topic_score_codex":0.010818622,"about_ca_topic_score_gemma":0.009412452,"teacher_disagreement_score":0.010818622,"about_ca_system_score_codex":0.002337301,"about_ca_system_score_gemma":0.002127219,"threshold_uncertainty_score":0.021511316},"labels":[],"label_agreement":null},{"id":"W2047310403","doi":"10.1007/s11219-012-9185-8","title":"Foreword to the special issue on quality engineering for software product lines","year":2012,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Advanced Software Engineering Methodologies","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Athabasca University","funders":"Alberta Innovates - Technology Futures","keywords":"Software product line; Domain engineering; Software engineering; Software development; Software quality management; Social software engineering; Computer science; Software construction; Software quality control; Product engineering; Software quality analyst; Software quality; Product (mathematics); Systems engineering; Engineering; Software; Product design","score_opus":0.09735351601825751,"score_gpt":0.37962411800898177,"score_spread":0.2822706019907243,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2047310403","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00020972223,0.015073599,0.0010534737,0.052589424,0.915201,0.00004684912,0.00018699587,0.00012044886,0.0155185675],"genre_scores_gemma":[0.0018229041,0.015516706,0.00078136724,0.032674376,0.7887506,0.00006919755,0.000537055,0.0003393891,0.15950835],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9985266,0.0001720268,0.00017683856,0.00019472232,0.0007994304,0.00013029946],"domain_scores_gemma":[0.9893176,0.0029666252,0.0008236567,0.00043394926,0.005176618,0.0012814391],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0019184385,0.002179461,0.0024541765,0.003146763,0.0018786651,0.0064000143,0.0016019554,0.006730598,0.08757013],"category_scores_gemma":[0.008544694,0.0005897214,0.001076817,0.0019925446,0.00090072764,0.004901199,0.0016348789,0.00713334,0.053727552],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000018836348,0.000015832595,0.000038020757,0.00013786375,0.000006149602,0.000034691468,0.000007460545,0.000023809807,0.00015323037,0.00054035377,0.9860407,0.012983143],"study_design_scores_gemma":[0.000017056573,0.000052305324,0.0006316791,0.0003098966,0.000019597632,0.00011919262,0.000035224166,0.00016976072,0.00019644552,0.0019954545,0.99644,0.000013295153],"about_ca_topic_score_codex":0.0013134036,"about_ca_topic_score_gemma":0.002825375,"teacher_disagreement_score":0.08757013,"about_ca_system_score_codex":0.0013306538,"about_ca_system_score_gemma":0.0018730466,"threshold_uncertainty_score":0.2929513},"labels":[],"label_agreement":null},{"id":"W2055986279","doi":"10.1007/s11219-014-9233-7","title":"Prioritizing code-smells correction tasks using chemical reaction optimization","year":2014,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":59,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Code refactoring; Code smell; Computer science; Metaheuristic; Code (set theory); Search-based software engineering; Software; Software engineering; Software quality; Software design; Programming language; Artificial intelligence; Software development","score_opus":0.04201576491452648,"score_gpt":0.32948212129263316,"score_spread":0.2874663563781067,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2055986279","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2984577,0.0009590948,0.67238605,0.0009029301,0.00022360579,0.00055096246,0.00033869888,0.017880494,0.008300344],"genre_scores_gemma":[0.6929396,0.0002675905,0.30071342,0.00021876946,0.00004113133,0.00012853205,0.0005079143,0.0009476161,0.0042354744],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9987459,0.00025608373,0.00006394155,0.00024831758,0.00049705396,0.00018864665],"domain_scores_gemma":[0.99459594,0.0028722482,0.0006519823,0.00038824196,0.0012296156,0.00026196186],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011707608,0.0017215442,0.0012948822,0.0017491925,0.0006554086,0.001091673,0.001319401,0.00079605484,0.004201456],"category_scores_gemma":[0.0063091246,0.00049634784,0.0010013545,0.0010515422,0.0004282046,0.000886875,0.0009551197,0.0012277453,0.000756151],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013526626,0.0012682638,0.007773323,0.00078801933,0.00016013179,0.0002868393,0.00013253231,0.31089112,0.12829125,0.0029541936,0.005546606,0.540555],"study_design_scores_gemma":[0.000053147924,0.0003176613,0.0012466535,0.000013436439,0.000075924974,0.00004712347,0.00006427932,0.9481966,0.04677472,0.0017898458,0.0013975739,0.000023057371],"about_ca_topic_score_codex":0.0068268934,"about_ca_topic_score_gemma":0.012460351,"teacher_disagreement_score":0.0068268934,"about_ca_system_score_codex":0.00093014457,"about_ca_system_score_gemma":0.003988785,"threshold_uncertainty_score":0.014055252},"labels":[],"label_agreement":null},{"id":"W2056306711","doi":"10.1007/s11219-012-9180-0","title":"Influence of confirmation biases of developers on software quality: an empirical study","year":2012,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":47,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Computer science; Assertion; Empirical research; Software quality; Confirmation bias; Software metric; Software development; Cognitive bias; Context (archaeology); Empirical evidence; Software bug; Software; Metric (unit); Quality (philosophy); Data science; Cognition; Software engineering; Psychology; Social psychology; Statistics; Engineering; Operations management; Mathematics; Programming language","score_opus":0.14038679657632722,"score_gpt":0.43106620055480366,"score_spread":0.29067940397847647,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2056306711","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9980628,0.00020250598,0.0004278989,0.000110113724,0.000008624272,0.000018303346,0.000030059065,0.000012346502,0.0011274166],"genre_scores_gemma":[0.9994332,0.000055864944,0.00028981618,0.000029375291,0.000013840024,0.000008723443,0.000025371673,0.0000081857,0.00013572405],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.966521,0.019528765,0.002187338,0.0019552226,0.008443899,0.0013637702],"domain_scores_gemma":[0.18400733,0.7021711,0.06753559,0.01581276,0.025191758,0.005281414],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.029912265,0.00043352385,0.0005991871,0.0027812703,0.0010733722,0.0024013312,0.00093294354,0.0014922724,0.0023008196],"category_scores_gemma":[0.36258754,0.0003899301,0.00051750056,0.0017567246,0.0015492042,0.0019873732,0.0015390759,0.0015138282,0.0003405603],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011070311,0.0006016135,0.9707959,0.00014761266,0.00014698628,0.0003008158,0.004421147,0.00043081,0.0014686738,0.00042763434,0.0003713258,0.01978031],"study_design_scores_gemma":[0.00016658471,0.0011767464,0.9830056,0.00013428763,0.0004940955,0.00063135993,0.0046594357,0.0039187414,0.0034840615,0.0010508134,0.0012121798,0.00006604176],"about_ca_topic_score_codex":0.0041642347,"about_ca_topic_score_gemma":0.004319952,"teacher_disagreement_score":0.9700877,"about_ca_system_score_codex":0.0015610728,"about_ca_system_score_gemma":0.0031363023,"threshold_uncertainty_score":0.15819305},"labels":[],"label_agreement":null},{"id":"W2063462200","doi":"10.1007/s11219-012-9186-7","title":"Mining the impact of evolution categories on object-oriented metrics","year":2012,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Cohesion (chemistry); Software evolution; Software; Computer science; Recall; Software metric; Software development; Data mining; Software engineering; Software construction; Programming language; Psychology; Cognitive psychology","score_opus":0.04837848210346001,"score_gpt":0.35755459381978144,"score_spread":0.3091761117163214,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2063462200","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9885921,0.0009939127,0.0060300357,0.00022122521,0.000051599003,0.000050013994,0.0024461478,0.00024663316,0.0013682981],"genre_scores_gemma":[0.9837843,0.00032370508,0.008975265,0.00004280442,0.000045447894,0.000041284322,0.0061019133,0.00006591298,0.0006193097],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99437296,0.0012559787,0.0006737834,0.000795623,0.002546149,0.0003554268],"domain_scores_gemma":[0.961573,0.024614122,0.004624015,0.002111102,0.005821512,0.0012563185],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0032369094,0.00079440704,0.00075492996,0.014446359,0.0005727726,0.0015047437,0.0009240642,0.0008244275,0.000971784],"category_scores_gemma":[0.032217916,0.0002625728,0.001038044,0.009467813,0.0003550654,0.0021689464,0.0009711356,0.000905853,0.0004205107],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00035997655,0.0004382124,0.8133389,0.00031757716,0.00044086116,0.00027485337,0.00034929757,0.0035074647,0.006186729,0.00087265874,0.0028283934,0.1710851],"study_design_scores_gemma":[0.00005577201,0.0007373326,0.88721716,0.00016143722,0.0006648742,0.00092056853,0.0009772258,0.09049374,0.0073513254,0.004889012,0.006477442,0.000054085005],"about_ca_topic_score_codex":0.005688388,"about_ca_topic_score_gemma":0.012971412,"teacher_disagreement_score":0.9967631,"about_ca_system_score_codex":0.0006774647,"about_ca_system_score_gemma":0.0011323489,"threshold_uncertainty_score":0.017118573},"labels":[],"label_agreement":null},{"id":"W2064132582","doi":"10.1007/s11219-010-9124-5","title":"Identification and analysis of attributes and base measures within ISO 9126","year":2010,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure; Université du Québec à Montréal","funders":"","keywords":"Measure (data warehouse); Software; Identification (biology); Task (project management); Software quality; Quality (philosophy); Reliability engineering; Set (abstract data type); Base (topology); Engineering; Computer science; Process (computing); Data mining; Systems engineering; Mathematics; Software development; Programming language","score_opus":0.03959637261930754,"score_gpt":0.3236045379643921,"score_spread":0.28400816534508455,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064132582","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8245511,0.00092845043,0.16010831,0.0002534023,0.000078761026,0.0004937911,0.0021121886,0.0008387227,0.010635205],"genre_scores_gemma":[0.9062373,0.00022541113,0.0890246,0.00002525531,0.000017295211,0.000307313,0.003114473,0.000119169665,0.0009291978],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9879502,0.0016104956,0.0012948596,0.0006550555,0.0080818245,0.00040758343],"domain_scores_gemma":[0.959201,0.014012897,0.004352208,0.004103525,0.017680487,0.0006498202],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008421085,0.000489884,0.0007519263,0.009011422,0.00093374564,0.0030801878,0.0010218546,0.0005603817,0.00066501077],"category_scores_gemma":[0.036267344,0.00032583726,0.00074868585,0.008863719,0.00062796264,0.0023417461,0.0011098371,0.0010864856,0.0003993863],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009644648,0.0009694057,0.32551637,0.0008723681,0.00017549992,0.000362075,0.0033037283,0.014626837,0.032828357,0.030758012,0.0030344801,0.58658844],"study_design_scores_gemma":[0.000094825635,0.0023639651,0.68604505,0.0005607279,0.00061217346,0.000982059,0.004639295,0.16907193,0.070124,0.030349385,0.03494305,0.00021358836],"about_ca_topic_score_codex":0.0037436294,"about_ca_topic_score_gemma":0.002972967,"teacher_disagreement_score":0.009011422,"about_ca_system_score_codex":0.0016759863,"about_ca_system_score_gemma":0.0029284717,"threshold_uncertainty_score":0.044535458},"labels":[],"label_agreement":null},{"id":"W2065045338","doi":"10.1007/s11219-009-9081-z","title":"A new perspective on data homogeneity in software cost estimation: a study in the embedded systems domain","year":2009,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":41,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"National Research Council Canada","funders":"Türkiye Bilimsel ve Teknolojik Araştırma Kurumu","keywords":"Computer science; Software; Cost estimate; Domain (mathematical analysis); Estimation; Data mining; Empirical research; Perspective (graphical); Domain engineering; Software development; Cost driver; Software metric; Data science; Artificial intelligence; Software construction; Systems engineering; Engineering; Statistics","score_opus":0.08691323916499845,"score_gpt":0.4021292808525964,"score_spread":0.315216041687598,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065045338","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07567355,0.0031333596,0.91064966,0.00439895,0.000112367445,0.00007041935,0.0002366654,0.00009900932,0.005625949],"genre_scores_gemma":[0.8362548,0.0022218276,0.15698422,0.0011605258,0.000996528,0.000121682104,0.00033583058,0.00026155036,0.0016630775],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.97665673,0.01343635,0.0014511057,0.0033594512,0.0044629625,0.00063339307],"domain_scores_gemma":[0.6015313,0.36060327,0.01100359,0.016022509,0.009178263,0.0016610553],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023308506,0.0009850585,0.002392342,0.0047776657,0.0014323463,0.0063779443,0.0041009984,0.0027530484,0.003452657],"category_scores_gemma":[0.1435379,0.001289496,0.002749367,0.0073946863,0.005193875,0.015940515,0.0047270497,0.0037895776,0.0002953201],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00060091534,0.00044251667,0.049145866,0.0009823862,0.00076705636,0.0011944465,0.003748137,0.077098824,0.0040404405,0.7068805,0.0024355054,0.1526634],"study_design_scores_gemma":[0.00009664695,0.0006341134,0.017428517,0.00031281257,0.0005039385,0.0011951191,0.001780729,0.35717613,0.004465568,0.60754377,0.008707846,0.00015476916],"about_ca_topic_score_codex":0.0036582558,"about_ca_topic_score_gemma":0.0016158228,"teacher_disagreement_score":0.023308506,"about_ca_system_score_codex":0.0021811237,"about_ca_system_score_gemma":0.0014429237,"threshold_uncertainty_score":0.123268604},"labels":[],"label_agreement":null},{"id":"W2069963480","doi":"10.1007/s11219-005-4253-y","title":"Virtual Software Engineering Laboratories in Support of Trade-off Analyses","year":2005,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":17,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"Bundesministerium für Bildung und Forschung","keywords":"Product (mathematics); Computer science; Software engineering; Software; Quality (philosophy); Software quality; Systems engineering; Empirical research; Software quality control; Engineering; Software development; Process management; Risk analysis (engineering); Business; Operating system","score_opus":0.04235531741497253,"score_gpt":0.3502672804459467,"score_spread":0.3079119630309741,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2069963480","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.35214695,0.0003564356,0.5999127,0.0012455212,0.00021873666,0.00023264285,0.00023429627,0.0110116955,0.03464103],"genre_scores_gemma":[0.92695373,0.00004803187,0.070484824,0.000053153763,0.000053617267,0.000103432016,0.00013483159,0.00018652437,0.0019818288],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99434245,0.0035422125,0.00018698975,0.000559806,0.0010066483,0.00036184493],"domain_scores_gemma":[0.943393,0.03302082,0.003345401,0.011878735,0.004930351,0.0034317658],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008525589,0.0006364626,0.00075634633,0.002085861,0.0012129009,0.0045494833,0.003464938,0.0014429181,0.010745098],"category_scores_gemma":[0.03770426,0.0005367282,0.0003035256,0.0017057372,0.0009963528,0.005674752,0.0048225857,0.0012650603,0.0014301178],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010404404,0.0024856133,0.029309237,0.00025126943,0.0002182032,0.0009707464,0.0017154763,0.18887661,0.022949431,0.12215649,0.013294006,0.6073685],"study_design_scores_gemma":[0.0010205741,0.0010971609,0.004030733,0.00005437948,0.00013828246,0.0003703275,0.00053077197,0.9105879,0.016724626,0.051912446,0.013443767,0.00008902187],"about_ca_topic_score_codex":0.0008143728,"about_ca_topic_score_gemma":0.0010707853,"teacher_disagreement_score":0.010745098,"about_ca_system_score_codex":0.0011046577,"about_ca_system_score_gemma":0.002080726,"threshold_uncertainty_score":0.045088172},"labels":[],"label_agreement":null},{"id":"W2073102918","doi":"10.1007/s11219-014-9238-2","title":"Studying the relationship between source code quality and mobile platform dependence","year":2014,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":45,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal; Queen's University","funders":"","keywords":"Android (operating system); Source code; Software quality; Computer science; Software quality assurance; Software; Operating system; Source lines of code; Mobile apps; Mobile device; Mobile computing; Quality assurance; Software development; World Wide Web; Engineering","score_opus":0.152923259708518,"score_gpt":0.38289060231190514,"score_spread":0.22996734260338714,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2073102918","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9986835,0.00009024021,0.0005867184,0.00007360908,0.0000018585881,0.0000032191897,0.000039048475,0.0000092741275,0.0005125532],"genre_scores_gemma":[0.99926704,0.000042560998,0.0002982651,0.000011350882,0.000004290886,0.0000023605492,0.00006449439,0.000009105524,0.0003005766],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99849224,0.00052598654,0.000113224065,0.00020837908,0.00047123723,0.00018899485],"domain_scores_gemma":[0.8319708,0.12849906,0.02551204,0.0035698,0.007920404,0.0025278763],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0023792926,0.0002692452,0.00019720233,0.001788352,0.0003628783,0.0010079637,0.0005978967,0.00060429773,0.002727931],"category_scores_gemma":[0.058522403,0.00031867484,0.00045559986,0.0020926571,0.00055852573,0.0019449948,0.0006875952,0.0015391022,0.0003291913],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017810545,0.00017820043,0.9900961,0.00002297915,0.000105799394,0.00006419946,0.00021334228,0.00091970916,0.0013213976,0.00038992046,0.00008833768,0.0064219525],"study_design_scores_gemma":[0.000008326422,0.00020952946,0.9890274,0.000013100275,0.000103263286,0.00012770019,0.00039433673,0.00822966,0.0011543201,0.00052035507,0.00020243901,0.000009542185],"about_ca_topic_score_codex":0.007842065,"about_ca_topic_score_gemma":0.013678543,"teacher_disagreement_score":0.007842065,"about_ca_system_score_codex":0.00067027385,"about_ca_system_score_gemma":0.00085211714,"threshold_uncertainty_score":0.0155928135},"labels":[],"label_agreement":null},{"id":"W2073803340","doi":"10.1007/s11219-007-9029-0","title":"In search for a widely applicable and accepted software quality model for software quality engineering","year":2007,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":46,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure; Ubisoft (Canada)","funders":"","keywords":"Software quality control; Software quality; Software quality analyst; Software quality management; Quality (philosophy); Software engineering; Computer science; Software development; Social software engineering; Software construction; Verification and validation; Software; Engineering; Operations management","score_opus":0.08279210101335747,"score_gpt":0.39928907863032326,"score_spread":0.31649697761696577,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2073803340","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014632359,0.0018996996,0.9591489,0.0103406,0.0003894887,0.00029553496,0.0001855286,0.0008238567,0.012284048],"genre_scores_gemma":[0.33093253,0.0027764812,0.6546943,0.0034599137,0.0005494322,0.0012109816,0.000580171,0.0005057073,0.005290471],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9658817,0.009208837,0.0028554278,0.0025123965,0.01840678,0.0011349135],"domain_scores_gemma":[0.92821944,0.025954992,0.005557071,0.009909133,0.029007407,0.0013519125],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.024937993,0.0014477514,0.00236735,0.0057241884,0.0018615643,0.008790161,0.0056382664,0.0049830284,0.0026435116],"category_scores_gemma":[0.09294712,0.00079850317,0.002361701,0.004099312,0.0035524347,0.014540286,0.0043043247,0.008476999,0.0015588907],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008588075,0.0004502345,0.006718341,0.0014086914,0.0002570779,0.00014255842,0.0012989955,0.030595524,0.0031752542,0.7353061,0.0120785255,0.20848294],"study_design_scores_gemma":[0.00009297553,0.00072971627,0.004980634,0.0013698769,0.00026673617,0.00035341838,0.0014455778,0.19085206,0.004201805,0.7476255,0.04794877,0.00013294959],"about_ca_topic_score_codex":0.0040269834,"about_ca_topic_score_gemma":0.0054892795,"teacher_disagreement_score":0.024937993,"about_ca_system_score_codex":0.0077683465,"about_ca_system_score_gemma":0.01099605,"threshold_uncertainty_score":0.13188624},"labels":[],"label_agreement":null},{"id":"W2077174673","doi":"10.1007/s11219-010-9112-9","title":"A comparative study for estimating software development effort intervals","year":2010,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":29,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"National Research Council Canada; Türkiye Bilimsel ve Teknolojik Araştırma Kurumu","keywords":"Computer science; Estimation; Software; Interval estimation; Interval (graph theory); Point (geometry); Point estimation; Data mining; Cluster (spacecraft); Machine learning; Statistics; Confidence interval; Engineering; Mathematics","score_opus":0.08125202920087467,"score_gpt":0.3986317212269463,"score_spread":0.31737969202607164,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2077174673","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8570223,0.0032463672,0.13231188,0.00007489122,0.000049432612,0.00022718974,0.0005885473,0.00034177015,0.006137593],"genre_scores_gemma":[0.907927,0.0004216958,0.09008731,0.000017055421,0.000021125144,0.000085565734,0.00059534126,0.00006429385,0.00078070967],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9823245,0.011314342,0.0017251077,0.0012390033,0.0030876002,0.00030935995],"domain_scores_gemma":[0.6082958,0.35717964,0.005749344,0.010553656,0.017396199,0.0008253861],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.024084039,0.0007910591,0.0011770193,0.008103463,0.0007761088,0.0018429015,0.0023866347,0.001204038,0.003584798],"category_scores_gemma":[0.15775347,0.0005251797,0.0012649032,0.0072573437,0.00047876206,0.0042481576,0.0009976261,0.0008029966,0.00044354115],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.012341227,0.001704809,0.2894871,0.0018189024,0.0012222183,0.00043180786,0.0036556164,0.043287847,0.012639932,0.0054777954,0.0009231398,0.6270097],"study_design_scores_gemma":[0.0008760198,0.021891912,0.42607144,0.00044678198,0.0032509011,0.0021174229,0.004476967,0.5037619,0.027077904,0.0052604964,0.004472729,0.00029544966],"about_ca_topic_score_codex":0.006643042,"about_ca_topic_score_gemma":0.0054913466,"teacher_disagreement_score":0.024084039,"about_ca_system_score_codex":0.0015719907,"about_ca_system_score_gemma":0.00069190277,"threshold_uncertainty_score":0.12737012},"labels":[],"label_agreement":null},{"id":"W2098513273","doi":"10.1007/s11219-006-5997-8","title":"PEM: The small company-dedicated software process quality evaluation method combining CMMISM and ISO/IEC 14598","year":2006,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":21,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure; Motorola (Canada); Computer Research Institute of Montréal","funders":"","keywords":"Capability Maturity Model Integration; LeanCMMI; Software quality control; Software quality; Personal software process; Software quality management; Computer science; Software engineering; Software quality analyst; Team software process; Software; Software development process; Verification and validation; Process (computing); Quality (philosophy); Software development; Systems engineering; Engineering; Software construction; Operations management","score_opus":0.07220551126872021,"score_gpt":0.3888549613939644,"score_spread":0.3166494501252442,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098513273","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.056773454,0.0003253982,0.91700995,0.0002475773,0.00008051982,0.0010961384,0.000822372,0.0061675194,0.017477097],"genre_scores_gemma":[0.29361707,0.00019964406,0.69613993,0.000106046034,0.000021348906,0.0010271139,0.0009082081,0.00037345104,0.007607173],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9964449,0.0012694423,0.00021631445,0.00034651707,0.0016293481,0.00009349088],"domain_scores_gemma":[0.99476373,0.0020315163,0.00047419418,0.00064292795,0.0019459779,0.00014159975],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0039247377,0.0008029413,0.0005905007,0.002729178,0.00041671054,0.0011886497,0.00079332944,0.00063784287,0.0058040232],"category_scores_gemma":[0.014456709,0.00033200582,0.0005222786,0.0016152735,0.0003620704,0.0024067597,0.0014847727,0.00052453583,0.0012823952],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00060139655,0.0002969229,0.011323931,0.0010033915,0.00013514675,0.00011549454,0.00085461733,0.003933098,0.022782218,0.009107916,0.005125832,0.94472],"study_design_scores_gemma":[0.0009133802,0.006256338,0.13109207,0.0012816224,0.0013893364,0.0031419718,0.0040719155,0.33083332,0.28169435,0.04439523,0.19457863,0.00035184238],"about_ca_topic_score_codex":0.0010621086,"about_ca_topic_score_gemma":0.0018803591,"teacher_disagreement_score":0.0058040232,"about_ca_system_score_codex":0.00076314976,"about_ca_system_score_gemma":0.0015056983,"threshold_uncertainty_score":0.020756245},"labels":[],"label_agreement":null},{"id":"W2106743089","doi":"10.1007/s11219-004-5259-6","title":"The Evolution Path for Industrial Software Quality Evaluation Methods Applying ISO/IEC 9126:2001 Quality Model: Example of MITRE?s SQAE Method","year":2005,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":26,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure","funders":"","keywords":"Software quality; Reliability engineering; Quality (philosophy); Software; Computer science; Verification and validation; Software quality control; Engineering; Systems engineering; Software development; Operations management; Operating system","score_opus":0.25675761791988566,"score_gpt":0.4826588867655611,"score_spread":0.22590126884567546,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2106743089","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03651257,0.0005780924,0.930137,0.0020505197,0.00007654877,0.00024824403,0.00016125702,0.0007027837,0.029532928],"genre_scores_gemma":[0.2096852,0.00041611484,0.7802092,0.00016772283,0.0000141640285,0.00024829514,0.0002396073,0.00024056114,0.0087791355],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9943295,0.0022710068,0.00032481272,0.00046984226,0.002416732,0.0001881188],"domain_scores_gemma":[0.98516315,0.00565581,0.0010679147,0.0017447666,0.006124237,0.00024410333],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0067314296,0.00045289315,0.00038995798,0.0024058507,0.0013458194,0.002155534,0.0010179252,0.001560072,0.00541616],"category_scores_gemma":[0.027442312,0.00052004267,0.00061481027,0.002222874,0.0014014115,0.0033685523,0.0015602629,0.0018842726,0.0015586651],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012891451,0.00026084672,0.012689386,0.00039212586,0.00005391497,0.00019715523,0.0024980567,0.021247815,0.007764296,0.41019228,0.005371984,0.5392033],"study_design_scores_gemma":[0.0001134366,0.0006764407,0.015134714,0.0007646364,0.00011747424,0.0009927524,0.0014528729,0.5889753,0.017522154,0.28398144,0.09012905,0.00013974887],"about_ca_topic_score_codex":0.0055378242,"about_ca_topic_score_gemma":0.006301778,"teacher_disagreement_score":0.0067314296,"about_ca_system_score_codex":0.0022594472,"about_ca_system_score_gemma":0.0026703316,"threshold_uncertainty_score":0.03559965},"labels":[],"label_agreement":null},{"id":"W2129182946","doi":"10.1007/s11219-007-9030-7","title":"Empirical studies to assess the understandability of data warehouse schemas using structural metrics","year":2007,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":54,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Computer Research Institute of Montréal; Université du Québec à Montréal","funders":"","keywords":"Computer science; Data mining; Metric (unit); Schema (genetic algorithms); Data warehouse; Similarity (geometry); Empirical research; Machine learning; Artificial intelligence; Statistics; Mathematics","score_opus":0.584142689272458,"score_gpt":0.5376677499733624,"score_spread":0.04647493929909563,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2129182946","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9971788,0.00011085559,0.0019188529,0.000066835135,0.0000031454,0.00006662227,0.000085018866,0.000012177549,0.00055779103],"genre_scores_gemma":[0.994705,0.00011240127,0.004469878,0.0000292937,0.000007247713,0.000075645,0.0004031491,0.000015745227,0.00018148597],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9783499,0.012099016,0.0032311694,0.0011236101,0.004687166,0.00050907873],"domain_scores_gemma":[0.35785428,0.57176244,0.031242324,0.012028494,0.025599087,0.0015134396],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.032877073,0.00055067195,0.0003442327,0.0039899335,0.00060045853,0.0015486703,0.0010766997,0.0010495356,0.0018656303],"category_scores_gemma":[0.26569083,0.00040672428,0.0007660653,0.004052996,0.0011400941,0.005919001,0.0010268776,0.0013581081,0.00024666215],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014029389,0.0061164177,0.90134746,0.0009957543,0.00057258626,0.00025676063,0.01747535,0.0036525235,0.0069306633,0.0027365454,0.00068218634,0.057830807],"study_design_scores_gemma":[0.00054807164,0.010917376,0.8933903,0.0005430387,0.0007444517,0.001103127,0.021806985,0.043310843,0.020101167,0.0035078556,0.0039140442,0.00011269851],"about_ca_topic_score_codex":0.0028477125,"about_ca_topic_score_gemma":0.0037708986,"teacher_disagreement_score":0.032877073,"about_ca_system_score_codex":0.0013482698,"about_ca_system_score_gemma":0.001245253,"threshold_uncertainty_score":0.17387265},"labels":[],"label_agreement":null},{"id":"W2139435518","doi":"10.1007/s11219-007-9015-6","title":"Introduction to the special issue on: “Software Quality Improvements and Estimations with Intelligence-based Methods”","year":2007,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Computer science; Software quality; Quality (philosophy); Software engineering; Software; Software development; Programming language","score_opus":0.0479755089915145,"score_gpt":0.39931576881835573,"score_spread":0.35134025982684125,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2139435518","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00089013804,0.044713646,0.04696248,0.034650303,0.83909845,0.00018881907,0.0006694189,0.0007321283,0.032094695],"genre_scores_gemma":[0.0064245537,0.040626433,0.015337167,0.024982845,0.79600865,0.00024510417,0.0018734016,0.0011264335,0.113375485],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99756646,0.0006156349,0.00034946427,0.00037898243,0.0009795127,0.000109845576],"domain_scores_gemma":[0.9890844,0.0051997183,0.0005342616,0.0006555425,0.0038924206,0.0006337173],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002417604,0.0021703814,0.002628638,0.0038975263,0.000985472,0.004180586,0.0020213316,0.003768035,0.05417098],"category_scores_gemma":[0.012641133,0.0007333988,0.0017041537,0.002607588,0.0010777372,0.004654613,0.0020147525,0.004654896,0.03288099],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000039054874,0.000035768124,0.00013065581,0.00041273132,0.000028772827,0.00007272854,0.000025319414,0.00021244575,0.0004576491,0.0029817333,0.9407628,0.054840237],"study_design_scores_gemma":[0.00002244094,0.00009196706,0.000938574,0.00030733054,0.000045652425,0.00031130068,0.00004091949,0.0010744218,0.00047230843,0.0074244947,0.9892273,0.00004323163],"about_ca_topic_score_codex":0.00065058086,"about_ca_topic_score_gemma":0.001401218,"teacher_disagreement_score":0.05417098,"about_ca_system_score_codex":0.00083839195,"about_ca_system_score_gemma":0.0010417441,"threshold_uncertainty_score":0.18122},"labels":[],"label_agreement":null},{"id":"W2308674594","doi":"10.1007/s11219-015-9304-4","title":"Efficient and scalable omniscient debugging for model transformations","year":2016,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Model-Driven Software Engineering Techniques","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"U.S. Department of Education","keywords":"Debugging; Computer science; Scalability; Algorithmic program debugging; Session (web analytics); Programming language; Software; Software engineering; Operating system","score_opus":0.030488092346449166,"score_gpt":0.2932194760941157,"score_spread":0.2627313837476665,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2308674594","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021968072,0.00019216577,0.9605309,0.00027439903,0.000052740514,0.00011533779,0.00016907815,0.01452708,0.002170229],"genre_scores_gemma":[0.40881604,0.0001630764,0.5862935,0.00012972837,0.000025273992,0.00010796015,0.0005883914,0.0019612485,0.0019147387],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9960316,0.0013790039,0.00024446155,0.0005731653,0.0014341136,0.00033760152],"domain_scores_gemma":[0.9891588,0.0049443664,0.0006078558,0.004030882,0.0010324535,0.00022560096],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028408612,0.0015749023,0.0013715448,0.0012886425,0.00092321454,0.0019909902,0.0032414664,0.0011701325,0.004145161],"category_scores_gemma":[0.017187,0.001055286,0.001129366,0.0012280412,0.0011592866,0.003930136,0.004015452,0.0020081725,0.0009676223],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012962798,0.0006514246,0.0058905296,0.0006891539,0.00022079554,0.0008153352,0.0006209384,0.20481932,0.040135927,0.0543055,0.017565612,0.6729892],"study_design_scores_gemma":[0.00014087553,0.00006185158,0.00033922217,0.000033669003,0.000045001027,0.00016079578,0.00007674097,0.9442671,0.013722204,0.03859047,0.0025382931,0.000023844908],"about_ca_topic_score_codex":0.0029609292,"about_ca_topic_score_gemma":0.0086228615,"teacher_disagreement_score":0.004145161,"about_ca_system_score_codex":0.0009795774,"about_ca_system_score_gemma":0.0024961638,"threshold_uncertainty_score":0.015024066},"labels":[],"label_agreement":null},{"id":"W2340362696","doi":"10.1007/s11219-016-9311-0","title":"A declarative framework for stateful analysis of execution traces","year":2016,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software System Performance and Reliability","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Computer science; Debugging; Tracing; TRACE (psycholinguistics); Programming language; Abstraction; Software engineering; Distributed computing; Theoretical computer science","score_opus":0.04390968862051765,"score_gpt":0.35937246303747283,"score_spread":0.3154627744169552,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2340362696","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007561829,0.00004602979,0.98977166,0.000062280196,0.000020121817,0.000075719094,0.00025547863,0.00866918,0.00034331344],"genre_scores_gemma":[0.07053434,0.00031947027,0.92136264,0.0002790562,0.00012665895,0.00046496818,0.0020271367,0.0032033946,0.0016823745],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9944634,0.0010488535,0.0008687982,0.00094909716,0.0021181605,0.000551663],"domain_scores_gemma":[0.98112124,0.009593331,0.0012873854,0.0051053343,0.0023915549,0.0005012101],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007703633,0.0024091732,0.0017631408,0.003551854,0.0015059012,0.007837732,0.007643584,0.0020586671,0.005893731],"category_scores_gemma":[0.021946289,0.0025042885,0.0041123326,0.002218544,0.0027871828,0.008054792,0.0053929863,0.005205796,0.0024916346],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00054556824,0.0005699995,0.0055530164,0.0015311773,0.00054685876,0.0013152226,0.0021624055,0.14439829,0.030111915,0.47166738,0.021011012,0.32058722],"study_design_scores_gemma":[0.00008794611,0.000104047904,0.0005714585,0.00030946097,0.00024351345,0.00041303982,0.00021356452,0.65405524,0.02431832,0.2904388,0.02909659,0.00014806195],"about_ca_topic_score_codex":0.0088071665,"about_ca_topic_score_gemma":0.011155307,"teacher_disagreement_score":0.0088071665,"about_ca_system_score_codex":0.0016300754,"about_ca_system_score_gemma":0.0046551474,"threshold_uncertainty_score":0.040741205},"labels":[],"label_agreement":null},{"id":"W2466590866","doi":"10.1007/s11219-016-9319-5","title":"Industry–academia collaborations in software testing: experience and success stories from Canada and Turkey","year":2016,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":46,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Hacettepe Üniversitesi; Türkiye Bilimsel ve Teknolojik Araştırma Kurumu","keywords":"Empirical research; Relevance (law); Confidentiality; Set (abstract data type); Software; Engineering; Engineering management; Knowledge management; Computer science; Software engineering; Business; Political science; Computer security","score_opus":0.05220059938723919,"score_gpt":0.31836247764262326,"score_spread":0.26616187825538407,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2466590866","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97498363,0.001584584,0.0005348134,0.0067608627,0.00014671787,0.00006964278,0.000051519863,0.00005659629,0.015811665],"genre_scores_gemma":[0.9940739,0.00033730065,0.00026715366,0.0005994145,0.000016116575,0.000013294446,0.000039486953,0.000028579672,0.0046247174],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.98487955,0.0046423795,0.00031021787,0.0008332032,0.0035049673,0.0058297827],"domain_scores_gemma":[0.93966603,0.01854504,0.0028428936,0.0011047231,0.012564787,0.02527649],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0098233,0.00047067946,0.00048046414,0.0027385121,0.027775459,0.0070228167,0.0024316635,0.0034429545,0.0041920617],"category_scores_gemma":[0.017990034,0.00039079477,0.00033658088,0.0047928626,0.00851196,0.002488824,0.009120047,0.0029483961,0.0005225507],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046739882,0.0014769912,0.1637893,0.00028370915,0.00007793443,0.009441895,0.6529613,0.00070272916,0.0024276804,0.010623283,0.022617647,0.13513018],"study_design_scores_gemma":[0.000054029573,0.00028478794,0.07359932,0.00023314891,0.000039530576,0.0011619952,0.8561633,0.000724024,0.00089637074,0.0008420464,0.06591248,0.00008884554],"about_ca_topic_score_codex":0.6356865,"about_ca_topic_score_gemma":0.8508813,"teacher_disagreement_score":0.36431348,"about_ca_system_score_codex":0.031109557,"about_ca_system_score_gemma":0.07622362,"threshold_uncertainty_score":0.732918},"labels":[],"label_agreement":null},{"id":"W2512805961","doi":"10.1007/s11219-016-9333-7","title":"Onboarding inexperienced developers: struggles and perceptions regarding automated testing","year":2016,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Onboarding; Perception; Software testing; Software engineering; Computer science; Engineering; Psychology; Software; Social psychology; Programming language","score_opus":0.04988978108270573,"score_gpt":0.33398380233613445,"score_spread":0.2840940212534287,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2512805961","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99453133,0.00022692127,0.00040734885,0.0014426918,0.00007429709,0.000024860126,0.000009551846,0.000025413332,0.003257543],"genre_scores_gemma":[0.9972012,0.00018345706,0.0004987322,0.00066548865,0.000027521877,0.0000128515785,0.000022584754,0.000026445381,0.00136172],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.98237634,0.0073757973,0.0010311363,0.0010457549,0.005932835,0.0022381647],"domain_scores_gemma":[0.83041793,0.08265093,0.026428275,0.003629261,0.020658555,0.03621501],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018472368,0.00051686855,0.0004911777,0.0019108724,0.0034194752,0.0059525757,0.001756539,0.0020128444,0.0032403558],"category_scores_gemma":[0.11093524,0.0006860877,0.00046207733,0.00083796144,0.002599213,0.0034092194,0.004825175,0.003027373,0.00043228024],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024325254,0.0011425376,0.44579044,0.00017905691,0.00008693,0.003653095,0.45756158,0.0002348037,0.003020766,0.00073303457,0.005630419,0.08172415],"study_design_scores_gemma":[0.000045232006,0.0009076236,0.30870587,0.0003201768,0.000068449175,0.002125717,0.6717669,0.0010101879,0.0007696664,0.0008317055,0.013346058,0.00010242146],"about_ca_topic_score_codex":0.010355482,"about_ca_topic_score_gemma":0.016930485,"teacher_disagreement_score":0.018472368,"about_ca_system_score_codex":0.0020523074,"about_ca_system_score_gemma":0.003804798,"threshold_uncertainty_score":0.09769243},"labels":[],"label_agreement":null},{"id":"W2521099350","doi":"10.1007/s11219-016-9340-8","title":"An empirical study on the effects of code visibility on program testability","year":2016,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Fundamental Research Funds for the Central Universities","keywords":"Testability; Visibility; Computer science; Code coverage; Artifact (error); Code (set theory); Software; Software quality; Software engineering; Software metric; Programming language; Software development; Reliability engineering; Engineering; Artificial intelligence; Set (abstract data type)","score_opus":0.06644119961895152,"score_gpt":0.4136435348790582,"score_spread":0.3472023352601067,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2521099350","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99830437,0.00013495814,0.00050927105,0.000049445378,0.0000031054062,0.000015194567,0.00004551435,0.000010242208,0.0009278771],"genre_scores_gemma":[0.999316,0.00004257713,0.00041934723,0.000010820271,0.0000042263737,0.000008163457,0.00004550693,0.0000073275082,0.00014620244],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9906233,0.0053901975,0.00068872934,0.0007281698,0.0020978444,0.00047171573],"domain_scores_gemma":[0.2573614,0.68324775,0.037909564,0.009921348,0.007607208,0.003952626],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00835709,0.00033466346,0.00030681907,0.0014439813,0.00047377165,0.0011034086,0.0007606927,0.00071892777,0.0032087197],"category_scores_gemma":[0.18259357,0.00034202874,0.00063265796,0.0013620445,0.0013440972,0.002047631,0.0011265568,0.0015645719,0.00020370493],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025687907,0.0044306405,0.9499137,0.00023190475,0.00025166085,0.00028038246,0.0027617882,0.0009941929,0.0034059407,0.0008441043,0.00027558784,0.034041267],"study_design_scores_gemma":[0.00008778366,0.0033089437,0.98966557,0.00004336519,0.00019960129,0.00031521736,0.0010882805,0.0025534031,0.0018670624,0.0003961188,0.00045524907,0.000019411727],"about_ca_topic_score_codex":0.0026062091,"about_ca_topic_score_gemma":0.0033874533,"teacher_disagreement_score":0.9916429,"about_ca_system_score_codex":0.0005622097,"about_ca_system_score_gemma":0.0010378022,"threshold_uncertainty_score":0.044197083},"labels":[],"label_agreement":null},{"id":"W2593117527","doi":"10.1007/s11219-017-9361-y","title":"An empirical study of crash-inducing commits in Mozilla Firefox","year":2017,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"Natural Sciences and Engineering Research Council of Canada; Fonds Québécois de la Recherche sur la Nature et les Technologies","keywords":"Crash; Computer science; Empirical research; Software engineering; Business; Programming language; Mathematics; Statistics","score_opus":0.09836511809479921,"score_gpt":0.4380468293221385,"score_spread":0.3396817112273393,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2593117527","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9991968,0.00002317268,0.000109584034,0.000058257727,0.000002845241,0.000013620101,0.000055761837,0.000008122755,0.00053183537],"genre_scores_gemma":[0.9991462,0.00002695476,0.00018611507,0.000025946181,0.000004530247,0.00001338903,0.00015950039,0.0000062662934,0.00043100462],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9978446,0.00077002397,0.00018073866,0.000239067,0.0007169374,0.00024866007],"domain_scores_gemma":[0.8925157,0.05750228,0.028372575,0.0038924539,0.013139288,0.004577668],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034020368,0.00023709146,0.00020468874,0.0018172882,0.001188373,0.0010274075,0.0008955069,0.0009777728,0.0029922812],"category_scores_gemma":[0.056456927,0.0003129886,0.00018602186,0.0012876488,0.0006934627,0.0012341834,0.0006467877,0.0018575585,0.00065236964],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003258003,0.0016009773,0.9776649,0.000054087628,0.000041061845,0.00038215367,0.005382492,0.00037446886,0.00092131674,0.00047085522,0.0010443819,0.011737539],"study_design_scores_gemma":[0.000028243134,0.00063745416,0.9861172,0.000038822312,0.00002730905,0.00022884959,0.008721125,0.0025291138,0.0004617306,0.00018926394,0.0010030622,0.000017889945],"about_ca_topic_score_codex":0.017659115,"about_ca_topic_score_gemma":0.031974994,"teacher_disagreement_score":0.017659115,"about_ca_system_score_codex":0.0012867214,"about_ca_system_score_gemma":0.0013738216,"threshold_uncertainty_score":0.03511268},"labels":[],"label_agreement":null},{"id":"W2626932320","doi":"10.1007/s11219-017-9375-5","title":"An investigation of the fault-proneness of clone evolutionary patterns","year":2017,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal; Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada","keywords":"clone (Java method); Cloning (programming); Software maintenance; Similarity (geometry); Biology; Genetics; Computer science; Software system; Software; Gene; Artificial intelligence; Programming language","score_opus":0.049937006655939524,"score_gpt":0.34101401970992457,"score_spread":0.29107701305398503,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2626932320","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99495536,0.00007864493,0.004406203,0.00002536024,0.000003267691,0.000013815474,0.0000676342,0.00005374885,0.00039582624],"genre_scores_gemma":[0.9964393,0.000032845164,0.0032199256,0.000004555181,0.0000031013578,0.000007360041,0.00012885622,0.000011104272,0.00015302027],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9987037,0.00034975258,0.00013297562,0.00027888775,0.00043365653,0.0001010449],"domain_scores_gemma":[0.93559545,0.046982154,0.008048328,0.0036469887,0.0050249794,0.0007020579],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0019203064,0.00021162935,0.0002516408,0.0025000884,0.0003889837,0.0007715722,0.0006280276,0.0006367077,0.0008391184],"category_scores_gemma":[0.035219852,0.00018753065,0.00032202885,0.0016214395,0.0005122606,0.0011305248,0.00040694512,0.00040210361,0.0001104377],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008323896,0.0003883156,0.8001522,0.00025163774,0.0002690602,0.0010898133,0.0018464165,0.028496874,0.042402513,0.0040138685,0.0004365912,0.11982047],"study_design_scores_gemma":[0.00006442275,0.0014837475,0.5945332,0.000068951784,0.00033334622,0.004442239,0.0013903914,0.3620661,0.02623709,0.0081421565,0.0011693354,0.000068995265],"about_ca_topic_score_codex":0.0009755567,"about_ca_topic_score_gemma":0.0011208368,"teacher_disagreement_score":0.0025000884,"about_ca_system_score_codex":0.00033721022,"about_ca_system_score_gemma":0.00033496224,"threshold_uncertainty_score":0.010155678},"labels":[],"label_agreement":null},{"id":"W2904898906","doi":"10.1007/s11219-018-9429-3","title":"FSM inference and checking sequence construction are two sides of the same coin","year":2018,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Machine Learning and Algorithms","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Computer Research Institute of Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"TRACE (psycholinguistics); Sequence (biology); Inference; Computer science; Nondeterministic algorithm; Finite-state machine; Model checking; Permutation (music); A priori and a posteriori; Algorithm; Automaton; Theoretical computer science; Artificial intelligence","score_opus":0.04786971646388845,"score_gpt":0.3511958778392323,"score_spread":0.30332616137534385,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2904898906","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0045860717,0.00081767136,0.9809368,0.0056484817,0.00054174504,0.00009753303,0.00013126219,0.0017001582,0.0055401917],"genre_scores_gemma":[0.2284017,0.0016501317,0.75345427,0.0032265505,0.0012478229,0.00030703028,0.00071219093,0.0020374202,0.008962896],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9653196,0.013063437,0.0025879731,0.0056026443,0.011384004,0.002042408],"domain_scores_gemma":[0.88485056,0.07288716,0.0028492678,0.030255377,0.007908899,0.00124881],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027472425,0.0014970105,0.003493237,0.0049590534,0.0028092742,0.008413745,0.006799051,0.0076670754,0.013881301],"category_scores_gemma":[0.102685414,0.002096594,0.0043773027,0.0039658076,0.0176089,0.028290724,0.012617432,0.009146781,0.0030878845],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000285092,0.000105375,0.0015374655,0.00043680143,0.00014543944,0.0002319147,0.00071011623,0.011388764,0.001899981,0.8192126,0.00738751,0.15665889],"study_design_scores_gemma":[0.000047105266,0.000044057433,0.00043420016,0.00019161307,0.00006685916,0.00022217078,0.00012705864,0.040146183,0.004228082,0.94442326,0.009983377,0.00008612751],"about_ca_topic_score_codex":0.0049215066,"about_ca_topic_score_gemma":0.0043164724,"teacher_disagreement_score":0.027472425,"about_ca_system_score_codex":0.003354397,"about_ca_system_score_gemma":0.0074881334,"threshold_uncertainty_score":0.14528978},"labels":[],"label_agreement":null},{"id":"W2911476493","doi":"10.1007/s11219-018-9428-4","title":"API trustworthiness: an ontological approach for software library adoption","year":2019,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University","funders":"","keywords":"Computer science; Trustworthiness; License; Software engineering; Quality (philosophy); Application programming interface; Software; World Wide Web; Data science; Computer security","score_opus":0.046976021699527935,"score_gpt":0.3156316965895039,"score_spread":0.268655674889976,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2911476493","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08397589,0.0005239139,0.8737751,0.004817682,0.00013675525,0.00034436182,0.0004238414,0.00058602844,0.035416357],"genre_scores_gemma":[0.7668827,0.00061476615,0.22703916,0.00038642174,0.000120280056,0.00028098884,0.0005863461,0.00022924795,0.0038599654],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99239606,0.002106194,0.0011136475,0.0009717308,0.0027314115,0.0006809585],"domain_scores_gemma":[0.98216337,0.0067702467,0.0020743306,0.004167372,0.0040921303,0.00073250534],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00840191,0.00060626515,0.0006520157,0.009464205,0.0037658429,0.007494507,0.0022829508,0.0020519071,0.002335867],"category_scores_gemma":[0.025968952,0.0009236011,0.002658297,0.00679422,0.00607375,0.017390586,0.0053547076,0.003571856,0.00053840346],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000034934135,0.00012518845,0.0133439265,0.00026148482,0.00010948286,0.00043625198,0.0092762485,0.0031775306,0.0028287505,0.9125018,0.0014622732,0.05644218],"study_design_scores_gemma":[0.00002443245,0.00010720567,0.015460284,0.0007016717,0.00053392025,0.0010921424,0.0122533,0.087410025,0.0052445866,0.8193418,0.05768638,0.00014423183],"about_ca_topic_score_codex":0.019207094,"about_ca_topic_score_gemma":0.015813293,"teacher_disagreement_score":0.019207094,"about_ca_system_score_codex":0.0046662055,"about_ca_system_score_gemma":0.0056302273,"threshold_uncertainty_score":0.04443413},"labels":[],"label_agreement":null},{"id":"W2913863212","doi":"10.1007/s11219-019-9440-3","title":"Fault model-driven testing from FSM with symbolic inputs","year":2019,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Computer Research Institute of Montréal","funders":"","keywords":"Finite-state machine; Test suite; Computer science; Symbolic execution; Implementation; Fault model; Fault injection; Fault (geology); Set (abstract data type); Model-based testing; Theoretical computer science; Fault coverage; Process (computing); Constraint (computer-aided design); Algorithm; Programming language; Test case; Software; Engineering; Machine learning","score_opus":0.044427028615976964,"score_gpt":0.29178708709886186,"score_spread":0.2473600584828849,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2913863212","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06583197,0.00009717899,0.9273753,0.00018220332,0.0000414407,0.000087534405,0.00027229945,0.0035991156,0.0025129616],"genre_scores_gemma":[0.88411176,0.000053073283,0.11372559,0.00007422734,0.000017973947,0.00011033166,0.00036562167,0.00027941624,0.001261923],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99827266,0.00047440734,0.00006605992,0.00019464477,0.00082975166,0.00016236525],"domain_scores_gemma":[0.9922501,0.006197876,0.00032577902,0.0006676059,0.00049315847,0.00006562087],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009337906,0.0010198068,0.00076089206,0.0012700465,0.0003353108,0.0007388509,0.0011179652,0.0009907731,0.003448348],"category_scores_gemma":[0.010744768,0.00033508873,0.00097275525,0.00067433,0.0010663408,0.0013799439,0.0012537916,0.0010940706,0.00035895436],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00051366753,0.00017310945,0.002097078,0.0003982434,0.0000826783,0.0009435636,0.00027805756,0.8015964,0.02530546,0.05358531,0.0014786496,0.1135477],"study_design_scores_gemma":[0.000026525111,0.000052555944,0.00019791232,0.000022623764,0.000016010783,0.00007868625,0.000010771173,0.9637212,0.009146635,0.026365839,0.0003536806,0.0000075839184],"about_ca_topic_score_codex":0.0027509637,"about_ca_topic_score_gemma":0.0040679793,"teacher_disagreement_score":0.003448348,"about_ca_system_score_codex":0.0010641209,"about_ca_system_score_gemma":0.0011415764,"threshold_uncertainty_score":0.011535823},"labels":[],"label_agreement":null},{"id":"W2924144435","doi":"10.1007/s11219-019-09442-9","title":"A large-scale empirical study of code smells in JavaScript projects","year":2019,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":34,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Code smell; JavaScript; Computer science; Scripting language; Software quality; World Wide Web; Programming language; Software development; Software","score_opus":0.05973784110211131,"score_gpt":0.36947289628882407,"score_spread":0.30973505518671274,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2924144435","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9994722,0.000027586713,0.00016534975,0.000036361092,0.0000013186526,0.000008397681,0.00006851893,0.0000072897024,0.00021301907],"genre_scores_gemma":[0.9991885,0.000029990259,0.00032703968,0.000028249246,0.0000042717356,0.000018015266,0.00021293928,0.000009261621,0.00018174587],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9962049,0.0013733964,0.00033097397,0.00055049453,0.0012299521,0.00031023176],"domain_scores_gemma":[0.86367196,0.082127884,0.030408308,0.006348079,0.01174616,0.005697606],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0053270487,0.0002494265,0.00026097224,0.0018875576,0.00078747346,0.0011913793,0.0006998732,0.0009205723,0.0009718423],"category_scores_gemma":[0.049172115,0.00035637847,0.0003603538,0.0023448062,0.0010650133,0.0027699368,0.0014195397,0.0015196716,0.0004385737],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013144236,0.001026372,0.9827628,0.000059534847,0.00006063784,0.00015842564,0.0046019903,0.00027111225,0.0010946674,0.00015127231,0.000451773,0.009230005],"study_design_scores_gemma":[0.000011728349,0.00026043106,0.99337935,0.00002631761,0.000016774597,0.00011310425,0.0037391833,0.0016226796,0.0003405706,0.00009471579,0.00038008785,0.000014864997],"about_ca_topic_score_codex":0.0063783554,"about_ca_topic_score_gemma":0.013891428,"teacher_disagreement_score":0.0063783554,"about_ca_system_score_codex":0.0008843442,"about_ca_system_score_gemma":0.0011715435,"threshold_uncertainty_score":0.028172433},"labels":[],"label_agreement":null},{"id":"W2925127136","doi":"10.1007/s11219-018-9437-3","title":"Testing self-healing cyber-physical systems under uncertainty: a fragility-oriented approach","year":2019,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":23,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"Horizon 2020 Framework Programme; Norges Forskningsråd","keywords":"Fragility; Reliability engineering; Normality; Computer science; Reliability (semiconductor); Cyber-physical system; Data mining; Engineering; Mathematics; Statistics","score_opus":0.04131213216160864,"score_gpt":0.308153895772097,"score_spread":0.2668417636104884,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2925127136","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030535502,0.0002591953,0.9653995,0.0007159987,0.000031532243,0.00008192324,0.00003354599,0.0002887724,0.0026541029],"genre_scores_gemma":[0.85621095,0.00034492958,0.14156246,0.0002353281,0.00008299107,0.00013495982,0.000059809063,0.0000936826,0.0012747884],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9965875,0.0014807327,0.00013919758,0.00038618344,0.0010867787,0.0003195131],"domain_scores_gemma":[0.9823162,0.013864918,0.0012994022,0.0011531773,0.0010794827,0.00028690003],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0047735344,0.0012648224,0.0012673995,0.0022469715,0.0007685533,0.0017754844,0.0025451365,0.001583158,0.0015324156],"category_scores_gemma":[0.01935368,0.0005434034,0.0013974147,0.0009732832,0.003906408,0.0031881293,0.0025541696,0.001810398,0.00012154525],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016207468,0.00014326042,0.0030979582,0.0002655426,0.00018235252,0.0004249565,0.0004655585,0.7956812,0.004424917,0.13741727,0.00057815167,0.057156794],"study_design_scores_gemma":[0.000012506993,0.00007731205,0.00026855065,0.000032868116,0.000040196348,0.00007704707,0.00009407955,0.9163173,0.0016120455,0.08112196,0.00033227136,0.000013904977],"about_ca_topic_score_codex":0.0025210597,"about_ca_topic_score_gemma":0.0017847456,"teacher_disagreement_score":0.0047735344,"about_ca_system_score_codex":0.0012433336,"about_ca_system_score_gemma":0.0014662859,"threshold_uncertainty_score":0.02524513},"labels":[],"label_agreement":null},{"id":"W2955136383","doi":"10.1007/s11219-019-09456-3","title":"Pieces of contextual information suitable for predicting co-changes? An empirical study","year":2019,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Commit; Computer science; Artifact (error); Metadata; Baseline (sea); Software; Software engineering; Set (abstract data type); Software development; Data science; Data mining; Information retrieval; Artificial intelligence; World Wide Web; Database; Programming language","score_opus":0.06126340530962837,"score_gpt":0.39334973887793756,"score_spread":0.3320863335683092,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2955136383","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99734586,0.0002058521,0.00096880493,0.000044616136,0.000008744604,0.000051914656,0.0004098354,0.000024045888,0.0009401875],"genre_scores_gemma":[0.9981856,0.00007318009,0.0010772878,0.000011435364,0.000009976766,0.000028656475,0.00048487366,0.000009651314,0.0001193395],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9963742,0.0016054336,0.00035612207,0.00056974805,0.0008587647,0.00023580404],"domain_scores_gemma":[0.90701705,0.06651692,0.01194282,0.006101497,0.0062066256,0.0022151412],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002956989,0.00038085107,0.0006928404,0.0042536883,0.0008013612,0.0016621368,0.0007876168,0.0010062845,0.0018849694],"category_scores_gemma":[0.044158652,0.00027131883,0.00047818647,0.0071565616,0.00057001435,0.0033078745,0.001344731,0.00082170335,0.0004357965],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010697294,0.00094444223,0.9487245,0.00020397958,0.00018057134,0.000347991,0.0027925754,0.0011515019,0.0015438335,0.00028948794,0.00041494827,0.042336393],"study_design_scores_gemma":[0.00003254128,0.0006482215,0.97727215,0.00007514185,0.0003097505,0.00031081963,0.0051734135,0.0119781615,0.0015292319,0.0006279981,0.0019988806,0.00004367818],"about_ca_topic_score_codex":0.0050307666,"about_ca_topic_score_gemma":0.007489387,"teacher_disagreement_score":0.0050307666,"about_ca_system_score_codex":0.00061167777,"about_ca_system_score_gemma":0.00065696775,"threshold_uncertainty_score":0.015638232},"labels":[],"label_agreement":null},{"id":"W3130726859","doi":"10.1007/s11219-020-09538-7","title":"An empirical study on the effect of community smells on bug prediction","year":2021,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Insect and Arachnid Ecology and Behavior","field":"Biochemistry, Genetics and Molecular Biology","cited_by":16,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Empirical research; Computer science; Computer security; Business; Statistics; Mathematics","score_opus":0.04319073875896441,"score_gpt":0.38051748767032306,"score_spread":0.33732674891135866,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3130726859","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99926144,0.00006196595,0.00025064926,0.00006030982,0.000001976052,0.000003880199,0.00006623194,0.0000056248045,0.00028793837],"genre_scores_gemma":[0.9994879,0.000027701039,0.00022739719,0.000014833661,0.000003827286,0.0000032874138,0.00007902052,0.00000426615,0.00015180155],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99675673,0.001945596,0.00025170547,0.00044437087,0.00043373957,0.00016797191],"domain_scores_gemma":[0.57520527,0.37688652,0.030689634,0.008402575,0.0050077694,0.0038081808],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006454652,0.00037167515,0.00031204053,0.0010697315,0.0004827425,0.00092333806,0.0005903091,0.0008582911,0.0033448932],"category_scores_gemma":[0.088000216,0.0002803967,0.0006576731,0.0008991281,0.0009467238,0.001577815,0.000798857,0.0016086119,0.00042913135],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028181248,0.00032298808,0.9942964,0.000014549157,0.00008752869,0.00005837962,0.00017240363,0.00034023044,0.00035524895,0.000054854925,0.00006843617,0.0039472017],"study_design_scores_gemma":[0.000017904149,0.0005340247,0.99190676,0.000010435045,0.00007096791,0.00015655831,0.00035899045,0.006339224,0.00031263515,0.00015840346,0.00012283702,0.000011154639],"about_ca_topic_score_codex":0.0046115504,"about_ca_topic_score_gemma":0.007716923,"teacher_disagreement_score":0.006454652,"about_ca_system_score_codex":0.0004345203,"about_ca_system_score_gemma":0.0004597573,"threshold_uncertainty_score":0.03413582},"labels":[],"label_agreement":null},{"id":"W3149010935","doi":"10.1007/s11219-021-09552-3","title":"Reconciliation of scrum and the project management process of the ISO/IEC 29110 standard-Entry profile—an experimental evaluation through usability measures","year":2021,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École de Technologie Supérieure","funders":"","keywords":"Scrum; Agile software development; Usability; Process management; Process (computing); Engineering management; Engineering; Software development process; Best practice; Software; Computer science; Software development; Software engineering; Knowledge management; Management","score_opus":0.06787216461876273,"score_gpt":0.38606037768116114,"score_spread":0.3181882130623984,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3149010935","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98781675,0.00019982818,0.009518616,0.000038156206,0.000035231897,0.00071056926,0.000057676003,0.00014185788,0.0014812635],"genre_scores_gemma":[0.96435493,0.00021661821,0.03284889,0.000058754726,0.000022305949,0.00077359274,0.00028524507,0.00008329923,0.0013564525],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9777981,0.014733773,0.001541212,0.0009903589,0.0044764634,0.0004601503],"domain_scores_gemma":[0.90198505,0.066552654,0.008218279,0.010189743,0.011540403,0.0015138827],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021252213,0.0006085044,0.0008700267,0.0013760449,0.0009152588,0.001897731,0.0013202865,0.0010348462,0.0012646044],"category_scores_gemma":[0.0695022,0.00053909875,0.00053398265,0.0012975261,0.00094587496,0.0022732236,0.001850758,0.0011313629,0.00025914639],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.02921726,0.034532867,0.065609016,0.0047034305,0.0005116351,0.00033367253,0.035091788,0.012641106,0.06531322,0.005082707,0.002120635,0.7448427],"study_design_scores_gemma":[0.0047564395,0.29313695,0.4081982,0.0015814033,0.0021094861,0.0012542459,0.01870755,0.09993088,0.14734952,0.002947706,0.019320367,0.00070727366],"about_ca_topic_score_codex":0.0018682345,"about_ca_topic_score_gemma":0.0023852852,"teacher_disagreement_score":0.021252213,"about_ca_system_score_codex":0.0010389932,"about_ca_system_score_gemma":0.0023796095,"threshold_uncertainty_score":0.1123938},"labels":[],"label_agreement":null},{"id":"W4361275045","doi":"10.1007/s11219-023-09621-9","title":"Machine learning application development: practitioners’ insights","year":2023,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"Fonds de recherche du Québec – Nature et technologies; Natural Sciences and Engineering Research Council of Canada","keywords":"Best practice; Leverage (statistics); Computer science; Engineering management; Process (computing); Software development process; Software development; Quality (philosophy); Software; Knowledge management; Software engineering; Data science; Engineering; Artificial intelligence","score_opus":0.0405552949918816,"score_gpt":0.3248114528857969,"score_spread":0.2842561578939153,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4361275045","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1256034,0.03158413,0.16373943,0.5472318,0.0013643757,0.00009472644,0.00013683864,0.00034565103,0.12989968],"genre_scores_gemma":[0.9048167,0.019053675,0.044007555,0.019202244,0.0012849413,0.00008801453,0.000087576635,0.00012416749,0.011335108],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.9937441,0.0030052215,0.0002561725,0.0004860827,0.0021923606,0.0003160739],"domain_scores_gemma":[0.93944794,0.04315346,0.0009997446,0.0020713096,0.012425234,0.0019022131],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01790935,0.00038630015,0.00029625694,0.0025672794,0.0011554825,0.004738406,0.0011819782,0.0027773925,0.0044612503],"category_scores_gemma":[0.039952822,0.00031023848,0.00021273071,0.0018198427,0.0029635644,0.008907821,0.002518629,0.0049652806,0.00088392093],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012201308,0.0006159809,0.01568779,0.0010952855,0.000044302327,0.00071769155,0.030418277,0.0036047988,0.0033776027,0.4408398,0.07277441,0.43070215],"study_design_scores_gemma":[0.00009772564,0.00034098106,0.010643829,0.0024401096,0.00006204987,0.0016429134,0.051803984,0.030603547,0.0058462494,0.49496153,0.40147102,0.0000860203],"about_ca_topic_score_codex":0.0016018257,"about_ca_topic_score_gemma":0.0031430929,"teacher_disagreement_score":0.01790935,"about_ca_system_score_codex":0.0018176247,"about_ca_system_score_gemma":0.003185698,"threshold_uncertainty_score":0.09471482},"labels":[],"label_agreement":null},{"id":"W4392591292","doi":"10.1007/s11219-024-09663-7","title":"A comprehensive catalog of refactoring strategies to handle test smells in Java-based systems","year":2024,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"Instituto Nacional de Ciência e Tecnologia para Engenharia de Software; Fundação de Amparo à Pesquisa do Estado da Bahia; Conselho Nacional de Desenvolvimento Científico e Tecnológico; Coordenação de Aperfeiçoamento de Pessoal de Nível Superior","keywords":"Code refactoring; Java; Computer science; Programming language; Software engineering; Operating system; Software","score_opus":0.05475254564650548,"score_gpt":0.3527974695137458,"score_spread":0.2980449238672403,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392591292","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.058454152,0.26150396,0.51963115,0.0025399558,0.00082227733,0.0022949532,0.030978244,0.05681363,0.066961735],"genre_scores_gemma":[0.07562591,0.17632598,0.6590689,0.001532899,0.0003528671,0.0008959882,0.05712475,0.0062906714,0.022782002],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9967457,0.00036934743,0.0008684666,0.00028781008,0.0015472542,0.00018133581],"domain_scores_gemma":[0.9871975,0.006871643,0.0011493817,0.0020897354,0.002410618,0.00028109015],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003068578,0.002052074,0.0015500144,0.014133684,0.0007599841,0.0021834613,0.0024836278,0.0014744595,0.008487599],"category_scores_gemma":[0.012936805,0.0012478057,0.0020948946,0.011089885,0.00031090973,0.003157371,0.0011475611,0.00097265927,0.004832172],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007899675,0.000220956,0.0013385419,0.0044590817,0.00011916102,0.00013012618,0.00013404075,0.0017336083,0.008317524,0.0012304853,0.015640195,0.9665973],"study_design_scores_gemma":[0.0003475355,0.0014024233,0.029491214,0.018802388,0.002452952,0.004868132,0.00048794082,0.045278095,0.11491415,0.016011128,0.76524425,0.0006998083],"about_ca_topic_score_codex":0.0026762711,"about_ca_topic_score_gemma":0.006799267,"teacher_disagreement_score":0.014133684,"about_ca_system_score_codex":0.0005071242,"about_ca_system_score_gemma":0.002319932,"threshold_uncertainty_score":0.028393865},"labels":[],"label_agreement":null},{"id":"W4413024210","doi":"10.1007/s11219-025-09727-2","title":"Investigating adversarial attacks in software analytics via machine learning explainability","year":2025,"lang":"en","type":"article","venue":"Software Quality Journal","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Software analytics; Analytics; Computer science; Adversarial system; Software; Artificial intelligence; Software engineering; Machine learning; Data science; Software development; Software construction; Operating system","score_opus":0.026464954080557494,"score_gpt":0.32756341787785526,"score_spread":0.30109846379729777,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413024210","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13152589,0.00059758243,0.86014384,0.0029456438,0.000081744656,0.00006268376,0.0000905702,0.000383833,0.004168288],"genre_scores_gemma":[0.98524946,0.00022640568,0.013294132,0.0001512074,0.00008493039,0.000030361807,0.000054990087,0.00005360346,0.0008549686],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9945893,0.0029983623,0.00013635973,0.0007044217,0.0011646565,0.00040686785],"domain_scores_gemma":[0.8699252,0.1156108,0.0053896885,0.006437523,0.0019300838,0.00070664636],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008069549,0.0009777383,0.0012192519,0.0012745204,0.0005872029,0.0020604988,0.001406371,0.0021141889,0.0019949367],"category_scores_gemma":[0.07300747,0.0005732108,0.0009361084,0.00081187714,0.0038707494,0.0050041364,0.0035574192,0.004106654,0.00014306874],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019279218,0.00009549213,0.0035502953,0.00011373123,0.00014650359,0.0001581612,0.00018821484,0.7579965,0.0018947272,0.2184549,0.0010816477,0.016127039],"study_design_scores_gemma":[0.000007247924,0.000028750668,0.00035799522,0.000011371928,0.0000113154465,0.000021124055,0.0000164554,0.9275478,0.0004924031,0.07131509,0.00018278776,0.000007604123],"about_ca_topic_score_codex":0.0014849607,"about_ca_topic_score_gemma":0.00084096874,"teacher_disagreement_score":0.008069549,"about_ca_system_score_codex":0.001652225,"about_ca_system_score_gemma":0.0011175084,"threshold_uncertainty_score":0.04267633},"labels":[],"label_agreement":null}]}