{"meta":{"query_hash":"38f6b9bf663c","filters":{"venue":"Proceedings of the 44th International Conference on Software Engineering"},"cohort_total":10,"direct_labels_cover":0,"predictions_cover":10,"exported":10,"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/38f6b9bf663c","api":"https://metacan.xera.ac/api/v1/cohort?venue=Proceedings+of+the+44th+International+Conference+on+Software+Engineering"},"results":[{"id":"W4284663063","doi":"10.1145/3510003.3510131","title":"ShellFusion","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software System Performance and Reliability","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Computer science; Shell (structure); Unix; Task (project management); Operating system; Programming language; World Wide Web; Software; Engineering","score_opus":0.015022719302810815,"score_gpt":0.22067473674403582,"score_spread":0.205652017441225,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284663063","genre_codex":"software","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":[0.0069396757,0.0014580899,0.19293576,0.0011366527,0.0016893087,0.00033652733,0.024123656,0.63158584,0.13979441],"genre_scores_gemma":[0.09210661,0.0025066736,0.19243371,0.004805475,0.0005374768,0.0010232057,0.22723179,0.27907383,0.20028123],"study_design_codex":"not_applicable","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99853706,0.00010759712,0.000109839995,0.0003555707,0.0006649124,0.00022494028],"domain_scores_gemma":[0.99859434,0.00022909419,0.00007313315,0.00053330953,0.00042574308,0.00014450568],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0013723066,0.0016203776,0.001504599,0.0010402263,0.0013847214,0.0035637026,0.004643589,0.0022943486,0.1518514],"category_scores_gemma":[0.005420384,0.00131024,0.0017857865,0.0009684146,0.0007488985,0.0062800655,0.0060956464,0.002949769,0.14511445],"study_design_candidate":"bench_or_experimental","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":[0.00052689767,0.00009395335,0.0015523349,0.0005744034,0.0001059029,0.00038455645,0.00028265076,0.0017250533,0.013371286,0.016387334,0.8911051,0.073890574],"study_design_scores_gemma":[0.00012424028,0.000074352574,0.0013652864,0.0001229277,0.000054488526,0.00066324865,0.00010248353,0.017285012,0.027192857,0.015902042,0.93700063,0.000112518144],"about_ca_topic_score_codex":0.0028608674,"about_ca_topic_score_gemma":0.0025562237,"teacher_disagreement_score":0.1518514,"about_ca_system_score_codex":0.0007828383,"about_ca_system_score_gemma":0.0013212324,"threshold_uncertainty_score":0.5079936},"labels":[],"label_agreement":null},{"id":"W4284670329","doi":"10.1145/3510003.3510076","title":"Social science theories in software engineering research","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Social software engineering; Software; Software Engineering Process Group; Computer science; Software development; Management science; Engineering ethics; Field (mathematics); Epistemology; Software development process; Software construction; Engineering; Mathematics","score_opus":0.047533180329221374,"score_gpt":0.3234322068789773,"score_spread":0.2758990265497559,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284670329","genre_codex":"other","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07895622,0.09571316,0.2613403,0.14211728,0.0022008498,0.0007022428,0.0004267425,0.00034432893,0.4181988],"genre_scores_gemma":[0.86423606,0.058785234,0.051127847,0.009675725,0.0026270677,0.0007981271,0.00029407887,0.00014306177,0.01231286],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9781953,0.013416261,0.00094241684,0.0014956932,0.005168413,0.0007818776],"domain_scores_gemma":[0.8602081,0.12373627,0.0056718285,0.0030158828,0.005549036,0.0018188863],"candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.025313722,0.0011511463,0.0012602552,0.015753567,0.0041864817,0.0099148955,0.0019300596,0.0043211333,0.0064039635],"category_scores_gemma":[0.04176472,0.00065945025,0.0012515481,0.014564938,0.021211484,0.013932861,0.0050655324,0.0054000807,0.0011981792],"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":[0.000011135649,0.00007831446,0.0039016139,0.00049522746,0.00007402335,0.00018916537,0.005809262,0.0027057116,0.00007502395,0.94862515,0.0036626575,0.034372702],"study_design_scores_gemma":[0.000016001588,0.000030085097,0.0019339854,0.00057204685,0.000022103786,0.00010441533,0.003082539,0.0023726427,0.00011855414,0.9495999,0.042121895,0.000025876572],"about_ca_topic_score_codex":0.0041887867,"about_ca_topic_score_gemma":0.0033574002,"teacher_disagreement_score":0.9958135,"about_ca_system_score_codex":0.0074325437,"about_ca_system_score_gemma":0.006741716,"threshold_uncertainty_score":0.13387334},"labels":[],"label_agreement":null},{"id":"W4284670745","doi":"10.1145/3510003.3510115","title":"Inferring and applying type changes","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University","funders":"Natural Sciences and Engineering Research Council of Canada; University of Colorado Boulder; National Science Foundation","keywords":"Code refactoring; Computer science; Plug-in; Programming language; Maintainability; Type (biology); Precision and recall; Data type; Software engineering; Information retrieval; Data mining; Artificial intelligence; Software","score_opus":0.03354994806902248,"score_gpt":0.2610165799266521,"score_spread":0.22746663185762964,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284670745","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":[0.11875535,0.0017718216,0.77815324,0.0013564508,0.00089068786,0.0010438057,0.011548524,0.0778802,0.008599901],"genre_scores_gemma":[0.25200614,0.0014139264,0.7063318,0.0010847972,0.00029912114,0.0004163557,0.023822686,0.00884079,0.0057845167],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.980243,0.003322518,0.0018975331,0.005345596,0.00833675,0.0008546785],"domain_scores_gemma":[0.94936156,0.025407739,0.003925497,0.0116526205,0.009105073,0.0005475162],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008309321,0.0021833614,0.001580775,0.0071194125,0.0014245707,0.0037277017,0.0030834766,0.0021674216,0.0027903002],"category_scores_gemma":[0.07982732,0.0017780826,0.0027453743,0.0030610221,0.0013624883,0.003858324,0.003251539,0.003722255,0.0025566234],"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":[0.00046119982,0.0003979106,0.12168821,0.002044215,0.0005081052,0.002383844,0.0029288104,0.024560085,0.031875174,0.011951217,0.043692213,0.75750905],"study_design_scores_gemma":[0.0002049811,0.00038757533,0.047225583,0.0011317536,0.0011447427,0.0041060154,0.0019489543,0.52663296,0.14563434,0.044941653,0.22614527,0.0004962073],"about_ca_topic_score_codex":0.010089878,"about_ca_topic_score_gemma":0.016876254,"teacher_disagreement_score":0.010089878,"about_ca_system_score_codex":0.0013839047,"about_ca_system_score_gemma":0.0044272407,"threshold_uncertainty_score":0.04394442},"labels":[],"label_agreement":null},{"id":"W4284672859","doi":"10.1145/3510003.3510105","title":"A grounded theory of coordination in remote-first and hybrid software teams","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":57,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Scrum; Team software process; Team effectiveness; Grounded theory; Knowledge management; Software development; Software; Distrust; Computer science; Process management; Engineering; Software development process; Psychology; Qualitative research; Sociology","score_opus":0.017391982887572346,"score_gpt":0.23651286103868208,"score_spread":0.21912087815110973,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284672859","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":[0.075833715,0.0010558371,0.77178484,0.011098559,0.0002216397,0.0010286381,0.00020078146,0.00018221191,0.13859391],"genre_scores_gemma":[0.8350246,0.0005945244,0.15981686,0.00056224054,0.000045261495,0.0009391526,0.00011775979,0.000036612477,0.0028629578],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.99135697,0.0058456035,0.00033710775,0.0007885245,0.0011671601,0.0005045335],"domain_scores_gemma":[0.9938863,0.0039526857,0.00052940496,0.00056449964,0.0005009041,0.0005662401],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0075549386,0.0005941245,0.0006334456,0.0026320056,0.0031073105,0.0060878657,0.0024040774,0.0019616184,0.0022628382],"category_scores_gemma":[0.0059534726,0.000508549,0.000986769,0.0020487069,0.018605625,0.0061858855,0.0037884184,0.002547171,0.000311937],"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":[0.000019295769,0.000060866434,0.0009956604,0.00014537103,0.000014170622,0.00016632097,0.025968378,0.0029321078,0.0003357364,0.95869577,0.00069598254,0.009970421],"study_design_scores_gemma":[0.00012823689,0.000117660165,0.0015399415,0.00045167268,0.000023207362,0.00024317524,0.022068515,0.013052941,0.00042133196,0.92629695,0.03562159,0.00003477866],"about_ca_topic_score_codex":0.006312552,"about_ca_topic_score_gemma":0.0047921776,"teacher_disagreement_score":0.008193996,"about_ca_system_score_codex":0.008193996,"about_ca_system_score_gemma":0.00675658,"threshold_uncertainty_score":0.059451878},"labels":[],"label_agreement":null},{"id":"W4284682233","doi":"10.1145/3510003.3510188","title":"Efficient online testing for DNN-enabled systems using surrogate-assisted and many-objective optimization","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":77,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; National Research Foundation of Korea; Fonds National de la Recherche Luxembourg; National Research Foundation; European Commission","keywords":"Computer science; Reliability (semiconductor); Fidelity; High fidelity; Reliability engineering; Software; Test strategy; Artificial neural network; Embedded system; Real-time computing; Distributed computing; Machine learning; Engineering; Operating system","score_opus":0.04979620041711109,"score_gpt":0.266732807629432,"score_spread":0.21693660721232094,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284682233","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":[0.10929246,0.0011185618,0.8798147,0.0008583917,0.000117766656,0.00017255131,0.00024090036,0.0026924212,0.005692321],"genre_scores_gemma":[0.87648654,0.00015071018,0.12028751,0.00026808053,0.000025060746,0.0001995299,0.00042879186,0.00026423737,0.0018894714],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99878603,0.0005264807,0.000063691296,0.00017253358,0.00025335772,0.00019788374],"domain_scores_gemma":[0.9929404,0.005540471,0.0004069603,0.00027644375,0.00058950426,0.0002462435],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002368916,0.0018167885,0.0015662586,0.00070412847,0.00040500137,0.0010248222,0.0018149767,0.0016607447,0.0034910997],"category_scores_gemma":[0.009252509,0.0008935453,0.0007874873,0.00038566228,0.0012534925,0.0014002003,0.0015804176,0.0022037935,0.00041001593],"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":[0.000093351286,0.000045414286,0.0005710026,0.00007457626,0.000024814472,0.000053769236,0.000014940208,0.983582,0.00057414756,0.0014686303,0.00039309182,0.013104329],"study_design_scores_gemma":[0.000005755069,0.000015573194,0.000034668574,0.0000045719526,0.0000020636433,0.0000045262827,0.000002764005,0.9986965,0.00017817308,0.0010046117,0.000049488914,0.0000014021743],"about_ca_topic_score_codex":0.0072391415,"about_ca_topic_score_gemma":0.009303014,"teacher_disagreement_score":0.0072391415,"about_ca_system_score_codex":0.0015508913,"about_ca_system_score_gemma":0.00208294,"threshold_uncertainty_score":0.014393985},"labels":[],"label_agreement":null},{"id":"W4284689801","doi":"10.1145/3510003.3510157","title":"Automated handling of anaphoric ambiguity in requirements","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software Engineering Research","field":"Computer Science","cited_by":52,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; Fonds National de la Recherche Luxembourg","keywords":"Ambiguity; Computer science; Anaphora (linguistics); Artificial intelligence; Natural language processing; Coreference; Natural language; Natural language understanding; Interpretation (philosophy); Ambiguity resolution; Automation; Machine translation; Feature (linguistics); Resolution (logic); Machine learning; Programming language; Linguistics","score_opus":0.0372657344313989,"score_gpt":0.28603588832004295,"score_spread":0.24877015388864404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284689801","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":[0.064432904,0.0004205688,0.91422963,0.0012781761,0.000039967996,0.00046788037,0.000686894,0.013083723,0.0053602555],"genre_scores_gemma":[0.3045946,0.00023179418,0.6903859,0.0003677434,0.000033610348,0.00019048313,0.0017696955,0.0006348733,0.001791301],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98780257,0.0057305316,0.0007252051,0.0014240202,0.0039877445,0.00032984573],"domain_scores_gemma":[0.97518456,0.014819437,0.0025468483,0.004091058,0.0031171916,0.00024096349],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005489571,0.0014268924,0.0008518225,0.0027365126,0.0010108805,0.0022479221,0.0016253161,0.0015284751,0.002427839],"category_scores_gemma":[0.032909725,0.0009327004,0.0015558329,0.0016402362,0.0010375951,0.0043901997,0.0031276585,0.0023574077,0.0014955065],"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":[0.00024319465,0.00041324602,0.0071193767,0.0012865567,0.0001578377,0.0009468944,0.0027216398,0.08308984,0.04058652,0.023101911,0.01537163,0.82496136],"study_design_scores_gemma":[0.000079320016,0.00018641725,0.0049907663,0.00020146114,0.00011507716,0.001671576,0.0016470724,0.82471997,0.05791468,0.06565583,0.042692646,0.00012525408],"about_ca_topic_score_codex":0.002268629,"about_ca_topic_score_gemma":0.003828986,"teacher_disagreement_score":0.005489571,"about_ca_system_score_codex":0.0010299924,"about_ca_system_score_gemma":0.0030125969,"threshold_uncertainty_score":0.029031992},"labels":[],"label_agreement":null},{"id":"W4284693098","doi":"10.1145/3510003.3510161","title":"Rotten apples spoil the bunch","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Advanced Malware Detection Techniques","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Malware; Computer science; Malware analysis; Android (operating system); Android malware; Control flow graph; Static analysis; Graph; Computer security; Operating system; Theoretical computer science; Programming language","score_opus":0.015371092592450997,"score_gpt":0.226527573737665,"score_spread":0.211156481145214,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284693098","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":[0.89222854,0.0054867514,0.02965245,0.0019043057,0.0007224938,0.00031303425,0.0060945135,0.011803445,0.05179439],"genre_scores_gemma":[0.94427884,0.0020923363,0.025299625,0.0009902305,0.00016440751,0.000066015076,0.005596157,0.0022699046,0.019242445],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99864095,0.00014790686,0.00006599655,0.00022579766,0.00071249163,0.0002068991],"domain_scores_gemma":[0.99666035,0.0012988176,0.0005070512,0.00064995675,0.00075120653,0.00013254116],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005654384,0.0011142185,0.00047833647,0.004019045,0.001405446,0.0015637624,0.00040401056,0.00082949316,0.0024458508],"category_scores_gemma":[0.0041530696,0.00043097284,0.00065817026,0.0014948694,0.0011867435,0.0030133089,0.0011543124,0.0012548681,0.002033074],"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":[0.0018350031,0.0003514974,0.2167602,0.0015037198,0.00035081367,0.0056650224,0.007297006,0.009140367,0.05086354,0.02328822,0.09105058,0.5918941],"study_design_scores_gemma":[0.00006427855,0.00088788645,0.40220007,0.0016851142,0.0005662126,0.012738445,0.005732818,0.076248564,0.06534098,0.023937533,0.41018912,0.0004089961],"about_ca_topic_score_codex":0.00458626,"about_ca_topic_score_gemma":0.009343541,"teacher_disagreement_score":0.00458626,"about_ca_system_score_codex":0.00051881705,"about_ca_system_score_gemma":0.00072165014,"threshold_uncertainty_score":0.009119093},"labels":[],"label_agreement":null},{"id":"W4284693828","doi":"10.1145/3510003.3510106","title":"Nessie","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; National Science Foundation","keywords":"Callback; Computer science; Asynchronous communication; Unit testing; Test case; Generator (circuit theory); JavaScript; Test (biology); Operating system; Programming language; Software; Machine learning; Computer network","score_opus":0.02496960565977006,"score_gpt":0.24090047315395033,"score_spread":0.21593086749418028,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284693828","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":[0.006846061,0.00035096772,0.9224019,0.00022568487,0.0001663431,0.00040137253,0.0015616381,0.048626043,0.019419959],"genre_scores_gemma":[0.08142891,0.00032544564,0.8785304,0.00041208035,0.000046902776,0.0006197538,0.0069701294,0.0074432436,0.024223166],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9973827,0.00062712363,0.00023139734,0.0005701202,0.0009875647,0.00020111245],"domain_scores_gemma":[0.9950222,0.0019246807,0.00036034748,0.0012873526,0.0012543979,0.00015109591],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0021166892,0.0014657374,0.00081524404,0.0019434107,0.00064628583,0.0017063366,0.0031761047,0.0012492231,0.028581657],"category_scores_gemma":[0.010058254,0.00079900096,0.0014057852,0.0009766886,0.0010226449,0.0024696775,0.0023476735,0.0017002036,0.014735238],"study_design_candidate":"bench_or_experimental","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":[0.0008457471,0.00035550585,0.0037481552,0.0013897289,0.00012501978,0.0004925689,0.00041825153,0.052025672,0.021915236,0.10462277,0.0651587,0.7489026],"study_design_scores_gemma":[0.00026890705,0.000530723,0.0011923198,0.00032844706,0.000100435806,0.001445174,0.0001589082,0.5655947,0.056740783,0.093885325,0.27963868,0.00011566629],"about_ca_topic_score_codex":0.0018355498,"about_ca_topic_score_gemma":0.0038669715,"teacher_disagreement_score":0.028581657,"about_ca_system_score_codex":0.0011152098,"about_ca_system_score_gemma":0.0019178286,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W4284697050","doi":"10.1145/3510003.3512765","title":"Bots for pull requests","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Mobile Crowdsensing and Crowdsourcing","field":"Computer Science","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University","funders":"Conselho Nacional de Desenvolvimento Científico e Tecnológico; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Coordenação de Aperfeiçoamento de Pessoal de Nível Superior; Universidade Tecnológica Federal do Paraná; National Science Foundation","keywords":"Computer science; Interface (matter); Human–computer interaction; Coding (social sciences); Participatory design; Information overload; Software; Interface design; User interface; World Wide Web; Engineering; Operating system","score_opus":0.021326146633776598,"score_gpt":0.2358149082008024,"score_spread":0.2144887615670258,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284697050","genre_codex":"methods","genre_gemma":"software","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"software","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19671823,0.00069734146,0.64111906,0.010420295,0.00060982053,0.0031491758,0.00048495585,0.012539196,0.13426198],"genre_scores_gemma":[0.7516643,0.00030422202,0.19554846,0.004079467,0.00013882374,0.0024865125,0.0004336445,0.0019470531,0.04339758],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9902318,0.0059351516,0.00040330482,0.00088616234,0.0018412826,0.0007022647],"domain_scores_gemma":[0.9795989,0.012018278,0.0014367057,0.0034642874,0.0025790192,0.00090277527],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009114162,0.0013062005,0.0005361095,0.0012688386,0.004219365,0.004022895,0.0013768944,0.0030416255,0.013848182],"category_scores_gemma":[0.02814126,0.0009378555,0.0009099048,0.0004595564,0.0029746324,0.008213752,0.0056654746,0.0025217254,0.003998939],"study_design_candidate":"bench_or_experimental","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":[0.0013847173,0.0009957264,0.022314869,0.0029596805,0.0001547752,0.0033763356,0.13192889,0.004851333,0.14560528,0.357963,0.06245177,0.26601368],"study_design_scores_gemma":[0.00032804324,0.0010285497,0.010975348,0.0012109817,0.00023986129,0.0027011945,0.028098602,0.058463387,0.035196118,0.1307988,0.730575,0.00038399023],"about_ca_topic_score_codex":0.0013856118,"about_ca_topic_score_gemma":0.0027657053,"teacher_disagreement_score":0.013848182,"about_ca_system_score_codex":0.0014574225,"about_ca_system_score_gemma":0.002402156,"threshold_uncertainty_score":0.048200846},"labels":[],"label_agreement":null},{"id":"W4284709654","doi":"10.1145/3510003.3510159","title":"CLEAR","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 44th International Conference on Software Engineering","topic":"Topic Modeling","field":"Computer Science","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Computer science; Task (project management); Natural language processing; Artificial intelligence; Word (group theory); Information retrieval; Embedding; Semantics (computer science); Programming language; Mathematics","score_opus":0.02487960478481576,"score_gpt":0.22987186236868118,"score_spread":0.2049922575838654,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284709654","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":[0.014455273,0.0062199133,0.4401936,0.004653733,0.0026834353,0.0010531719,0.039009,0.12514542,0.36658654],"genre_scores_gemma":[0.12521695,0.0051580835,0.26723787,0.003846587,0.0010379626,0.0007515481,0.12765297,0.02170447,0.44739357],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9975248,0.00031258428,0.00019325259,0.000733267,0.0010005955,0.00023560897],"domain_scores_gemma":[0.99562705,0.0005993117,0.00028156143,0.0017055698,0.0015943267,0.00019215237],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016898963,0.0016324151,0.0009828237,0.0030805082,0.0014880387,0.0043710913,0.0020604117,0.0016720807,0.12776339],"category_scores_gemma":[0.007465617,0.0007992721,0.0012783081,0.0026979272,0.0005984146,0.008078856,0.002925541,0.0017125119,0.16235392],"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":[0.0003591476,0.0001250791,0.0032143595,0.0009726884,0.00008075906,0.00033294107,0.00022432432,0.0017868285,0.0076345,0.037458234,0.4081528,0.5396584],"study_design_scores_gemma":[0.00003799892,0.00007034931,0.0028885708,0.00018950747,0.000054080905,0.00067241356,0.0001918831,0.014376384,0.008497684,0.02000208,0.9529409,0.000078076395],"about_ca_topic_score_codex":0.0050761895,"about_ca_topic_score_gemma":0.0067111193,"teacher_disagreement_score":0.12776339,"about_ca_system_score_codex":0.00075602817,"about_ca_system_score_gemma":0.0016005567,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null}]}