{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":996,"total_is_capped":false,"direct_labels_cover":2,"predictions_cover":996,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"100aa09bc2bb","filters":{"topic":"Software Testing and Debugging Techniques"}},"results":[{"id":"W1577404745","doi":"10.1007/s10009-007-0044-z","title":"The software model checker Blast","year":2007,"lang":"en","type":"article","venue":"International Journal on Software Tools for Technology Transfer","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":594,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Computer science; Predicate abstraction; Predicate (mathematical logic); Programming language; Memory safety; Model checking; Programmer; Program analysis; Symbolic execution; Static analysis; Software","authors":[{"name":"Dirk Beyer","is_ca":true},{"name":"Thomas A. Henzinger","is_ca":false},{"name":"Ranjit Jhala","is_ca":false},{"name":"Rupak Majumdar","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0292661240647697,"gpt":0.3019749021356067,"spread":0.272708778070837,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001538838,0.0009394044,0.001007763,0.001399866,0.0007671447,0.001874422,0.001727358,0.001523155,0.009388052],"category_scores_gemma":[0.009137308,0.000836429,0.0009589174,0.001022949,0.001138449,0.00361428,0.002033445,0.002298445,0.003462769],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000591718,"about_ca_system_score_gemma":0.002096661,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002664579,"about_ca_topic_score_gemma":0.003686569,"domain_scores_codex":[0.9980152,0.0005311157,0.0001103409,0.0003188596,0.000876713,0.0001478969],"domain_scores_gemma":[0.9967263,0.001288921,0.0001649566,0.001192518,0.0005339804,0.00009341149],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001095852,0.0003966122,0.004156827,0.0008025023,0.0002601248,0.0009579579,0.0003163709,0.108885,0.02958077,0.2737637,0.09733807,0.482446],"study_design_scores_gemma":[0.0004143246,0.0002353559,0.0007914242,0.0002753837,0.0002195651,0.000743638,0.00007372219,0.6032544,0.04506527,0.2504209,0.09844049,0.00006556918],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.03068473,0.000747805,0.8591639,0.00121939,0.0004702523,0.0001364047,0.001101446,0.07537647,0.03109955],"genre_scores_gemma":[0.4389973,0.0008296994,0.5260752,0.001138666,0.0001771471,0.000195605,0.003360122,0.00766588,0.02156033],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.009388052,"threshold_uncertainty_score":0.03140616,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2106072155","doi":"10.1145/2568225.2568271","title":"Coverage is not strongly correlated with test suite effectiveness","year":2014,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":398,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"Test suite; Suite; Computer science; Proxy (statistics); Test (biology); Confounding; Test case; Statistics; Data mining; Machine learning; Mathematics; Regression analysis; Geology; Geography","authors":[{"name":"Laura Inozemtseva","is_ca":true},{"name":"Reid Holmes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.008925411068723877,"gpt":0.22585647833184,"spread":0.2169310672631161,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01239843,0.0007505367,0.001197942,0.004840515,0.0003535021,0.00211816,0.001124419,0.001325452,0.003066745],"category_scores_gemma":[0.2781792,0.000629343,0.001124094,0.003668169,0.001480655,0.003488568,0.001583485,0.001864594,0.000964816],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009791044,"about_ca_system_score_gemma":0.0008200747,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002069494,"about_ca_topic_score_gemma":0.002060566,"domain_scores_codex":[0.9697632,0.008215898,0.003070915,0.004101339,0.01292686,0.001921789],"domain_scores_gemma":[0.4663228,0.4363829,0.05638976,0.01444244,0.02148316,0.004978943],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005141038,0.0001314062,0.9465784,0.0003841924,0.001149146,0.0003243153,0.0002999493,0.008851292,0.002352961,0.000982551,0.001413948,0.03701774],"study_design_scores_gemma":[0.00003195093,0.0005581657,0.9708583,0.0001473919,0.0004636299,0.001233998,0.0002240738,0.01794206,0.002527745,0.003978797,0.001980738,0.00005316856],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9579586,0.004813873,0.0217529,0.001251169,0.00009558294,0.0001033302,0.001735916,0.0007169973,0.01157165],"genre_scores_gemma":[0.9969276,0.0002541098,0.00136371,0.00009800543,0.000048491,0.00003046355,0.0008495739,0.0001221869,0.0003059609],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01239843,"threshold_uncertainty_score":0.06557,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4391974543","doi":"10.1109/tse.2024.3368208","title":"Software Testing With Large Language Models: Survey, Landscape, and Vision","year":2024,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":394,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"National Natural Science Foundation of China","keywords":"Computer science; Software engineering; Software","authors":[{"name":"Junjie Wang","is_ca":false},{"name":"Yuchao Huang","is_ca":false},{"name":"Chunyang Chen","is_ca":false},{"name":"Zhe Liu","is_ca":false},{"name":"Song Wang","is_ca":true},{"name":"Qing Wang","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01712040667441351,"gpt":0.2519216021540697,"spread":0.2348011954796562,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009385707,0.001491541,0.001848627,0.00599255,0.0004309003,0.004482325,0.003635035,0.002141713,0.002385037],"category_scores_gemma":[0.06181108,0.00100749,0.001519923,0.005038078,0.003055187,0.01052341,0.002492735,0.003093329,0.001103475],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002108319,"about_ca_system_score_gemma":0.00291645,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006331891,"about_ca_topic_score_gemma":0.004358084,"domain_scores_codex":[0.9900885,0.004911423,0.0006522069,0.001158091,0.002961443,0.0002282895],"domain_scores_gemma":[0.9138474,0.07639063,0.002021867,0.002982831,0.004228889,0.0005283364],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000145883,0.0002058172,0.01046892,0.003307177,0.0001709392,0.0001961956,0.0005721956,0.01650292,0.001008697,0.02595612,0.01321999,0.9282451],"study_design_scores_gemma":[0.0001181542,0.0008701161,0.01419034,0.00932717,0.0004632761,0.002972635,0.002998374,0.4816467,0.008588703,0.1844773,0.2940071,0.0003400746],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.03519735,0.4594508,0.4426527,0.03379083,0.0006956821,0.0003308217,0.000978464,0.00620898,0.02069431],"genre_scores_gemma":[0.4255341,0.3314767,0.2209292,0.008213248,0.003103178,0.0005827866,0.003668368,0.001898819,0.004593555],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.009385707,"threshold_uncertainty_score":0.04963696,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4367672983","doi":"10.1016/j.jss.2023.111734","title":"GitHub Copilot AI pair programmer: Asset or Liability?","year":2023,"lang":"en","type":"article","venue":"Journal of Systems and Software","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":340,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"York University; Polytechnique Montréal","funders":"Fonds de recherche du Québec; Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Programmer; Liability; Asset (computer security); Computer science; Computer security; Software engineering; Business; Programming language; Finance","authors":[{"name":"Arghavan Moradi Dakhel","is_ca":true},{"name":"Vahid Majdinasab","is_ca":true},{"name":"Amin Nikanjam","is_ca":true},{"name":"Foutse Khomh","is_ca":true},{"name":"Michel C. Desmarais","is_ca":true},{"name":"Zhen Ming Jiang","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0401045157013695,"gpt":0.3012254088637339,"spread":0.2611208931623644,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005709504,0.0009220233,0.0005815612,0.00146974,0.0007558225,0.002100422,0.001147361,0.001351339,0.6416379],"category_scores_gemma":[0.004786971,0.000455053,0.0003972314,0.0009450378,0.0005162688,0.002069074,0.002723379,0.001258546,0.4622111],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006686947,"about_ca_system_score_gemma":0.0008923124,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002032984,"about_ca_topic_score_gemma":0.004338275,"domain_scores_codex":[0.9996239,0.00005316303,0.00001414961,0.00005820109,0.0001819913,0.00006856156],"domain_scores_gemma":[0.9974061,0.0005254421,0.0001265647,0.0007384719,0.0005990751,0.0006044887],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009848455,0.00002552686,0.0002840387,0.00007677682,0.000004466433,0.0002135113,0.00003073284,0.0001249223,0.0006176147,0.002318288,0.8932247,0.1029809],"study_design_scores_gemma":[0.00006628327,0.00004811429,0.001412061,0.0001836651,0.00001034848,0.0008493924,0.0000820781,0.002472985,0.001447293,0.007649889,0.985753,0.00002509771],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.007218229,0.001249985,0.03623437,0.009350669,0.003354958,0.0003888304,0.01027449,0.09740206,0.8345264],"genre_scores_gemma":[0.02810739,0.001057703,0.01888845,0.003927327,0.001128152,0.0002107459,0.008820674,0.0277089,0.9101507],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6416379,"threshold_uncertainty_score":0.5111601,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3043761819","doi":"10.1145/3395363.3397369","title":"CoCoNuT: combining context-aware neural translation models using ensemble for program repair","year":2020,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":329,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Facebook","keywords":"Computer science; Context (archaeology); Machine translation; Artificial intelligence; Programming language; Machine learning; Software engineering; Natural language processing","authors":[{"name":"Thibaud Lutellier","is_ca":true},{"name":"Hung Viet Pham","is_ca":true},{"name":"Lawrence Pang","is_ca":true},{"name":"Yitong Li","is_ca":true},{"name":"Moshi Wei","is_ca":true},{"name":"Lin Tan","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1833401154001179,"gpt":0.3325084569739183,"spread":0.1491683415738004,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001403226,0.002429406,0.001866596,0.002025556,0.001050213,0.001718728,0.002550834,0.002208497,0.006176425],"category_scores_gemma":[0.004590666,0.0005311625,0.001279082,0.00216328,0.000472692,0.002730581,0.001985937,0.002282435,0.003824456],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009610077,"about_ca_system_score_gemma":0.00177298,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01933593,"about_ca_topic_score_gemma":0.03235472,"domain_scores_codex":[0.9989523,0.0002410393,0.00005854419,0.0004403686,0.0001976157,0.0001101105],"domain_scores_gemma":[0.9983,0.0007421717,0.00007720917,0.0002646797,0.0005346805,0.00008126622],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0007443565,0.0004676324,0.00255327,0.0004507138,0.000490029,0.0004950628,0.0001898541,0.1218722,0.0116761,0.002406065,0.03531808,0.8233366],"study_design_scores_gemma":[0.00003614831,0.0001032712,0.0003649644,0.00003358239,0.00010205,0.00009370947,0.0000538503,0.986635,0.003885145,0.004136001,0.004525202,0.00003099974],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07254525,0.007894713,0.8362793,0.001140376,0.001632969,0.0003880895,0.003520635,0.06595095,0.01064771],"genre_scores_gemma":[0.529006,0.002456016,0.4258242,0.001603377,0.0006690269,0.0006070838,0.01775899,0.0042372,0.01783811],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01933593,"threshold_uncertainty_score":0.03844678,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1536265389","doi":"10.1007/3-540-46423-9_2","title":"Optimizing Java Bytecode Using the Soot Framework: Is It Feasible?","year":2000,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":313,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bytecode; Java bytecode; Computer science; Java; Programming language; Java annotation; Class (philosophy); Generics in Java; Program optimization; Java applet; Operating system; Compiler; Artificial intelligence","authors":[{"name":"Raja Vallée-Rai","is_ca":true},{"name":"Étienne Gagnon","is_ca":true},{"name":"Laurie Hendren","is_ca":true},{"name":"Patrick Lam","is_ca":true},{"name":"Patrice Pominville","is_ca":true},{"name":"Vijay Sundaresan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05289934935660727,"gpt":0.3031957885134227,"spread":0.2502964391568154,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00115195,0.0009612363,0.001074208,0.0004554459,0.0004338395,0.001586468,0.001724411,0.000963838,0.005710925],"category_scores_gemma":[0.005253985,0.0005940288,0.0009196803,0.0007018499,0.000878086,0.004897524,0.0007802619,0.001545177,0.001876437],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000525958,"about_ca_system_score_gemma":0.001530004,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003691611,"about_ca_topic_score_gemma":0.005886809,"domain_scores_codex":[0.9986153,0.0003169234,0.00005612117,0.0001816018,0.0005206033,0.0003095644],"domain_scores_gemma":[0.9972768,0.0009661342,0.0001733229,0.0009676834,0.0005208286,0.00009518422],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009462987,0.0002950038,0.004262058,0.0005423957,0.0001749892,0.0001403301,0.0002016409,0.05462862,0.03080404,0.1070989,0.02414279,0.7767628],"study_design_scores_gemma":[0.0002590775,0.0004727968,0.003275601,0.000291982,0.0003147408,0.0003941937,0.0003251677,0.636728,0.07826398,0.2128261,0.06670976,0.0001386112],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1023994,0.003654298,0.8342586,0.00356082,0.0005343759,0.00008674145,0.0002055446,0.02677541,0.02852483],"genre_scores_gemma":[0.5190296,0.002088822,0.4590565,0.0006420853,0.000158703,0.00008102616,0.0005033676,0.006222447,0.01221735],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005710925,"threshold_uncertainty_score":0.01910496,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2025618033","doi":"10.1007/s10270-002-0004-8","title":"A UML-Based Approach to System Testing","year":2002,"lang":"en","type":"article","venue":"Software & Systems Modeling","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":303,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Computer science; Unified Modeling Language; Sequence diagram; Class diagram; Object Constraint Language; Software engineering; Applications of UML; Testability; Model-based testing; Programming language; UML tool; Context (archaeology); Test case; Test Management Approach; Systems Modeling Language; Activity diagram; System testing; Software development; Reliability engineering; Software; Software construction","authors":[{"name":"Lionel Briand","is_ca":true},{"name":"Yvan Labiche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09020794658056883,"gpt":0.2392422866175647,"spread":0.1490343400369959,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005754771,0.001140987,0.0008314478,0.003237966,0.0010371,0.00343138,0.003284051,0.001997009,0.005385592],"category_scores_gemma":[0.01883765,0.001087818,0.001647562,0.001489386,0.002281447,0.005301193,0.002524943,0.003775695,0.001603697],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001385985,"about_ca_system_score_gemma":0.001959346,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003171894,"about_ca_topic_score_gemma":0.00464994,"domain_scores_codex":[0.9917076,0.003501742,0.0005244516,0.0004615456,0.00353784,0.0002667792],"domain_scores_gemma":[0.9908714,0.004882849,0.0003701552,0.002089709,0.001560879,0.0002250476],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00007865132,0.0003291239,0.001054004,0.000337144,0.0001016066,0.000330659,0.0007536746,0.04209584,0.009260534,0.6227638,0.007445042,0.31545],"study_design_scores_gemma":[0.00008856354,0.0001477966,0.0006707059,0.0004470727,0.0001692706,0.000826291,0.0001557766,0.4362407,0.01347579,0.4791965,0.06850389,0.00007772334],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0008865838,0.000141517,0.9941871,0.0003244034,0.00004083728,0.00006556288,0.00002652638,0.0009066582,0.003420738],"genre_scores_gemma":[0.05121123,0.0003507326,0.9442317,0.0002958536,0.00007275666,0.0001947028,0.0001500489,0.0003355113,0.003157462],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005754771,"threshold_uncertainty_score":0.03043449,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3004493192","doi":"10.1016/j.jss.2020.110542","title":"On testing machine learning programs","year":2020,"lang":"en","type":"article","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":198,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Computer science; Transformative learning; Domain (mathematical analysis); Software testing; Software engineering; Test strategy; Field (mathematics); Software; Witness; Data science; Artificial intelligence; Machine learning; Psychology","authors":[{"name":"Houssem Ben Braiek","is_ca":true},{"name":"Foutse Khomh","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02772335947957308,"gpt":0.238025687806026,"spread":0.2103023283264529,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005438473,0.002412608,0.001689657,0.003910051,0.001144177,0.001782613,0.004615527,0.002585734,0.008551338],"category_scores_gemma":[0.06106046,0.001278825,0.00204793,0.003476035,0.003617962,0.00854791,0.004396002,0.003399719,0.001070421],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002187337,"about_ca_system_score_gemma":0.001854638,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008490797,"about_ca_topic_score_gemma":0.01206508,"domain_scores_codex":[0.9829312,0.007889721,0.0009614869,0.002269737,0.004868865,0.001078994],"domain_scores_gemma":[0.8986333,0.08192474,0.001594327,0.01248297,0.00449524,0.0008694676],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001033769,0.0006561562,0.01037528,0.0006831214,0.0003051084,0.0006826919,0.0004967625,0.1292358,0.009342832,0.121542,0.02142987,0.7042165],"study_design_scores_gemma":[0.0001266216,0.0002885397,0.002589931,0.0002325501,0.0001736887,0.0004388753,0.0001534572,0.6812806,0.01229817,0.2944688,0.007908997,0.00003979039],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.07285555,0.002247791,0.8975637,0.002961765,0.0003947412,0.0001768369,0.0003821117,0.006710056,0.01670742],"genre_scores_gemma":[0.607393,0.001010585,0.3744785,0.001334959,0.000487614,0.0002521727,0.001944003,0.001838176,0.01126105],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.008551338,"threshold_uncertainty_score":0.02876174,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1928770023","doi":"10.1109/icsm.1989.65194","title":"Insights into regression testing (software testing)","year":2003,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":184,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Regression testing; Testability; Test plan; Computer science; Non-regression testing; Test (biology); Plan (archaeology); Risk-based testing; Regression analysis; Test Management Approach; Manual testing; Test strategy; Set (abstract data type); Reliability engineering; Regression; Keyword-driven testing; White-box testing; Machine learning; Data mining; Measure (data warehouse); Software; Statistics; Programming language; Mathematics; Engineering; Software system","authors":[{"name":"Hareton Leung","is_ca":true},{"name":"Lee White","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04500660842911066,"gpt":0.2756500233323609,"spread":0.2306434149032503,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003438883,0.001581155,0.0008155059,0.003482362,0.0006659047,0.003657505,0.001474942,0.002042243,0.005940976],"category_scores_gemma":[0.01493326,0.000708757,0.001063582,0.002571878,0.005660327,0.005890639,0.001650686,0.003757933,0.00134888],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002309652,"about_ca_system_score_gemma":0.001262561,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003850762,"about_ca_topic_score_gemma":0.002309347,"domain_scores_codex":[0.9953956,0.002393882,0.0001906136,0.0004382765,0.001295595,0.0002860432],"domain_scores_gemma":[0.9889717,0.008752327,0.0005513321,0.0005622337,0.0009491758,0.0002133817],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00002503824,0.00006470463,0.0008292695,0.0002516378,0.00001799731,0.0002854518,0.0007070996,0.008701543,0.0006175257,0.9049387,0.005502894,0.07805812],"study_design_scores_gemma":[0.00001423948,0.00005444835,0.000762147,0.0001820954,0.00001387612,0.0004448132,0.0001936454,0.02417176,0.0004665254,0.9359081,0.0377664,0.00002195712],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.007694375,0.01376521,0.9036773,0.01102884,0.0003799142,0.00009022399,0.0001722735,0.0007150738,0.06247678],"genre_scores_gemma":[0.4013324,0.02032828,0.5446941,0.005422507,0.00163167,0.0004326497,0.000473752,0.0005485099,0.0251362],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005940976,"threshold_uncertainty_score":0.01987451,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2125727889","doi":"10.1145/1368088.1368136","title":"Sufficient mutation operators for measuring test effectiveness","year":2008,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":181,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Mutant; Mutation testing; Test suite; Mutation; Measure (data warehouse); Computer science; Process (computing); Mathematics; Test case; Data mining; Genetics; Biology; Machine learning; Programming language; Gene","authors":[{"name":"Akbar Siami Namin","is_ca":true},{"name":"James H. Andrews","is_ca":true},{"name":"Duncan J. Murdoch","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04459589575011752,"gpt":0.266663017612815,"spread":0.2220671218626974,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01333238,0.002005143,0.001997803,0.007645034,0.0006736141,0.001492531,0.001863577,0.002903197,0.002793981],"category_scores_gemma":[0.08885669,0.0005421343,0.001333758,0.00283067,0.002223791,0.004002893,0.001332631,0.001388716,0.0009614908],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001020524,"about_ca_system_score_gemma":0.001190352,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006913241,"about_ca_topic_score_gemma":0.0006543014,"domain_scores_codex":[0.9551332,0.01348586,0.004839275,0.003390349,0.0217487,0.00140261],"domain_scores_gemma":[0.8653868,0.1007373,0.012469,0.009414095,0.009944702,0.002048032],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00383034,0.001776652,0.1310738,0.002635005,0.001027775,0.001794962,0.0008616951,0.1995937,0.1944632,0.1229153,0.007316164,0.3327115],"study_design_scores_gemma":[0.0004508228,0.002797318,0.06105242,0.0005388951,0.0005194231,0.005045681,0.0003257138,0.6557104,0.1463473,0.1159014,0.01100059,0.0003101102],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3793491,0.003615242,0.6006094,0.0004897168,0.0001455988,0.0006364717,0.002781363,0.003531559,0.008841516],"genre_scores_gemma":[0.8574812,0.0005051683,0.137912,0.0001937135,0.0001547128,0.0007030097,0.002073559,0.0004415056,0.0005349718],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01333238,"threshold_uncertainty_score":0.0705092,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4385187421","doi":"10.1109/sp46215.2023.10179420","title":"Examining Zero-Shot Vulnerability Repair with Large Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":178,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Zero (linguistics); Computer science; Vulnerability (computing); Ground zero; Computer security; Physics; Linguistics; Nuclear physics","authors":[{"name":"Hammond Pearce","is_ca":false},{"name":"Benjamin Tan","is_ca":true},{"name":"Baleegh Ahmad","is_ca":false},{"name":"Ramesh Karri","is_ca":false},{"name":"Brendan Dolan-Gavitt","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07209621342546572,"gpt":0.3021107687630803,"spread":0.2300145553376146,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005121061,0.001026674,0.0005169607,0.0009871738,0.000526232,0.0009475536,0.001719511,0.001714379,0.001569238],"category_scores_gemma":[0.03573749,0.0005928908,0.000644773,0.0005223839,0.001446016,0.002413653,0.001383772,0.002540188,0.0005392524],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00135165,"about_ca_system_score_gemma":0.001220934,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008922956,"about_ca_topic_score_gemma":0.01482413,"domain_scores_codex":[0.9971963,0.001716107,0.00009787315,0.0005324847,0.0003317549,0.0001255063],"domain_scores_gemma":[0.9561964,0.03912226,0.00100659,0.00204218,0.001285277,0.0003471498],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006690684,0.0008243561,0.02155022,0.000676582,0.0002057136,0.0008983324,0.002166665,0.8499894,0.008743327,0.004981066,0.008630355,0.1006649],"study_design_scores_gemma":[0.00005899735,0.0002321429,0.001706123,0.00003171348,0.00002686867,0.0001408231,0.0003684779,0.9874246,0.004125804,0.004283469,0.001576205,0.00002491643],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8943498,0.0009576162,0.09484602,0.001384863,0.0001146449,0.0001686646,0.0008431315,0.004984567,0.002350893],"genre_scores_gemma":[0.9575008,0.0001400233,0.03909775,0.0002658395,0.00002463534,0.0001211778,0.001444122,0.000346886,0.001058828],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008922956,"threshold_uncertainty_score":0.0270831,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2092198121","doi":"10.1016/j.jss.2012.12.051","title":"A survey of software testing practices in Canada","year":2013,"lang":"en","type":"article","venue":"Journal of Systems and Software","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":168,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary","funders":"","keywords":"Software testing; Test (biology); Survey research; Software; Engineering; Computer science; Psychology; Applied psychology","authors":[{"name":"Vahid Garousi","is_ca":true},{"name":"Junji Zhi","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06050471719038154,"gpt":0.2706308278705806,"spread":0.2101261106801991,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002579038,0.0003822451,0.0005204394,0.009695053,0.004579158,0.003094968,0.002462995,0.0009267799,0.005292805],"category_scores_gemma":[0.0171491,0.0005227926,0.0004342439,0.0242846,0.001622847,0.001029139,0.001297126,0.0009334587,0.0005867861],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.06614102,"about_ca_system_score_gemma":0.1318221,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.989943,"about_ca_topic_score_gemma":0.9936674,"domain_scores_codex":[0.9928902,0.0004234004,0.0005505293,0.000663714,0.004125809,0.001346273],"domain_scores_gemma":[0.9423838,0.008392444,0.005499649,0.0006736644,0.03447647,0.008573928],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004400791,0.0002222024,0.6843159,0.001257032,0.0001176731,0.001021689,0.01159897,0.0008263279,0.001929114,0.003997481,0.03048984,0.2637837],"study_design_scores_gemma":[0.00002158615,0.0000999816,0.909713,0.0007945314,0.00005125896,0.000525748,0.01073579,0.0006151226,0.0005456188,0.0001799923,0.07664217,0.00007526838],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.871035,0.04697299,0.001885381,0.0114696,0.000224502,0.0003341478,0.01167778,0.0003420442,0.05605853],"genre_scores_gemma":[0.9514769,0.02710006,0.002217138,0.002401513,0.00003322876,0.00007698491,0.002688234,0.0001205698,0.01388523],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.997421,"threshold_uncertainty_score":0.479889,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2092245922","doi":"10.1016/j.infsof.2013.03.004","title":"Graphical user interface (GUI) testing: Systematic mapping and repository","year":2013,"lang":"en","type":"article","venue":"Information and Software Technology","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":167,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Graphical user interface testing; Graphical user interface; Computer science; User interface; Software; Interface (matter); Information retrieval; Data mining; Software engineering; User interface design; Programming language; Operating system","authors":[{"name":"Ishan Banerjee","is_ca":false},{"name":"Bao N. Nguyen","is_ca":false},{"name":"Vahid Garousi","is_ca":true},{"name":"Atif M. Memon","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01087614462494123,"gpt":0.2151209586309208,"spread":0.2042448140059796,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01209981,0.001548791,0.001180586,0.01240615,0.0014439,0.003815389,0.0038487,0.001544135,0.002708847],"category_scores_gemma":[0.07745802,0.0009618949,0.001277146,0.006200252,0.001791086,0.006727997,0.004184421,0.001509238,0.001396955],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001340431,"about_ca_system_score_gemma":0.007884588,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004841406,"about_ca_topic_score_gemma":0.008874613,"domain_scores_codex":[0.9694749,0.01570686,0.002477125,0.001983209,0.009592635,0.0007652069],"domain_scores_gemma":[0.9064861,0.04338626,0.006799906,0.01790901,0.02463614,0.0007826102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001702376,0.0007418906,0.02808614,0.002313496,0.0001459901,0.0005169958,0.003752435,0.005628964,0.01012959,0.01305329,0.004786159,0.9306749],"study_design_scores_gemma":[0.0003920992,0.004606061,0.09238295,0.009605181,0.001878255,0.01482543,0.01936777,0.2989326,0.2775752,0.1326999,0.147016,0.0007186669],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.07386839,0.002753717,0.9007135,0.0008015806,0.00006903624,0.001651856,0.001119523,0.007948518,0.01107388],"genre_scores_gemma":[0.3379145,0.00197447,0.6522949,0.0002743131,0.00003436696,0.0009039903,0.002122809,0.001129596,0.003351033],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.01240615,"threshold_uncertainty_score":0.06399071,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2136921053","doi":"10.1109/tse.2010.58","title":"The Effects of Time Constraints on Test Case Prioritization: A Series of Controlled Experiments","year":2010,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":166,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"Regression testing; Prioritization; Computer science; Context (archaeology); Risk-based testing; Reliability engineering; Process (computing); Software; Regression analysis; Data mining; Risk analysis (engineering); Machine learning; Software development; Engineering; Management science; Software construction","authors":[{"name":"Hyunsook Do","is_ca":false},{"name":"Siavash Mirarab","is_ca":false},{"name":"Ladan Tahvildari","is_ca":true},{"name":"Gregg Rothermel","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.005484845015205005,"gpt":0.2193811498363502,"spread":0.2138963048211452,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01476421,0.003478669,0.001762069,0.001234533,0.001312446,0.002148462,0.004320335,0.003055643,0.005864657],"category_scores_gemma":[0.06549779,0.001350492,0.001691876,0.001310148,0.002729094,0.00241166,0.001796083,0.004881511,0.0006025335],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002565466,"about_ca_system_score_gemma":0.002879321,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002435528,"about_ca_topic_score_gemma":0.002870249,"domain_scores_codex":[0.9861057,0.005807009,0.001875114,0.00271731,0.002204289,0.001290611],"domain_scores_gemma":[0.7893358,0.1733025,0.0181555,0.008236971,0.00777192,0.003197293],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"nonrandomized_trial","study_design_gemma":"nonrandomized_trial","study_design_scores_codex":[0.1816958,0.3312303,0.01693113,0.004962215,0.002253009,0.001112787,0.008107449,0.0459279,0.2677897,0.006374319,0.004996781,0.1286187],"study_design_scores_gemma":[0.06333205,0.6317912,0.04611544,0.0005238389,0.003022663,0.0003337327,0.001996083,0.05529432,0.1668949,0.01336118,0.0163853,0.0009493094],"study_design_candidate":"nonrandomized_trial","study_design_consensus":"nonrandomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9740252,0.00038227,0.01160327,0.0002065688,0.0003282995,0.01008417,0.0005800266,0.0002992407,0.00249113],"genre_scores_gemma":[0.872146,0.0006013354,0.06541961,0.001205285,0.0003370331,0.05410937,0.00113904,0.0002570871,0.004785241],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01476421,"threshold_uncertainty_score":0.07808149,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2045837563","doi":"10.1007/s10664-012-9219-7","title":"Static test case prioritization using topic models","year":2012,"lang":"en","type":"article","venue":"Empirical Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":156,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Test suite; Computer science; Code coverage; Test case; Source code; Black box; Test Management Approach; Programming language; White-box testing; Test (biology); Test data; Operating system; Software; Software development; Artificial intelligence; Machine learning","authors":[{"name":"Stephen W. Thomas","is_ca":true},{"name":"Hadi Hemmati","is_ca":true},{"name":"Ahmed E. Hassan","is_ca":true},{"name":"Dorothea Blostein","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06432137932201937,"gpt":0.3036697114581389,"spread":0.2393483321361196,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01008298,0.00173376,0.001728762,0.01111392,0.001078136,0.003485714,0.002431113,0.00192231,0.007127939],"category_scores_gemma":[0.06890455,0.001014186,0.002514977,0.005228688,0.0005267462,0.00470247,0.002073967,0.00204661,0.001583351],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001991152,"about_ca_system_score_gemma":0.003448137,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008395957,"about_ca_topic_score_gemma":0.01281657,"domain_scores_codex":[0.9887435,0.00588415,0.0008405673,0.001650548,0.002206416,0.0006749231],"domain_scores_gemma":[0.9297371,0.05603277,0.002119486,0.003859625,0.007100894,0.001150158],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002298456,0.001120871,0.07243875,0.001190008,0.0008860973,0.000497332,0.002896034,0.1088954,0.01246731,0.02268837,0.0172624,0.757359],"study_design_scores_gemma":[0.0002269482,0.000357278,0.01059447,0.00009439843,0.0006066588,0.0003376809,0.0006551347,0.9425092,0.007078066,0.03263905,0.004808185,0.00009302979],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1531271,0.0008959026,0.8264585,0.0008056418,0.0001358418,0.0008818822,0.001583622,0.009680969,0.006430659],"genre_scores_gemma":[0.7826588,0.00029061,0.2088859,0.0001286524,0.0001307931,0.0007405065,0.002977544,0.001102442,0.003084827],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01111392,"threshold_uncertainty_score":0.05332452,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2116368564","doi":"10.1145/1985793.1985870","title":"Automated cross-browser compatibility testing","year":2011,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":147,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Compatibility (geochemistry); Computer science; Web browser; Client-side scripting; Emulation; World Wide Web; Backward compatibility; Web-based simulation; Pairwise comparison; Web navigation; Software engineering; Information retrieval; Web page; The Internet; Web API; Operating system; Engineering; Artificial intelligence","authors":[{"name":"Ali Mesbah","is_ca":true},{"name":"Mukul R. Prasad","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1194683359133998,"gpt":0.3234254279076075,"spread":0.2039570919942077,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00441626,0.0009229603,0.0008002445,0.001422805,0.0006892781,0.001119692,0.0019233,0.001238647,0.001985722],"category_scores_gemma":[0.0190355,0.0006565145,0.001119166,0.0006660812,0.001088508,0.002140318,0.002947199,0.001168184,0.0004663657],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006908028,"about_ca_system_score_gemma":0.001750076,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002393965,"about_ca_topic_score_gemma":0.003262701,"domain_scores_codex":[0.9898399,0.004941123,0.0004981855,0.001265228,0.002795293,0.0006602988],"domain_scores_gemma":[0.975055,0.01475586,0.001724189,0.00568153,0.002481407,0.0003019901],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0009472959,0.001889695,0.07713678,0.0006070428,0.0003687777,0.002733086,0.002921162,0.2550527,0.122043,0.04709502,0.005031619,0.4841737],"study_design_scores_gemma":[0.00006305955,0.0002799291,0.005220468,0.0000465263,0.00006548907,0.0005499489,0.0002416004,0.9196927,0.04600369,0.02533199,0.002458873,0.00004576118],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2371152,0.000098132,0.7491394,0.0001798271,0.00003425213,0.0002397858,0.0001912244,0.01033869,0.002663375],"genre_scores_gemma":[0.8175883,0.00005045582,0.1802338,0.00007234509,0.00001098029,0.0001310849,0.0004926056,0.0004807363,0.0009397947],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.00441626,"threshold_uncertainty_score":0.02335566,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2158798798","doi":"10.1145/1368088.1368118","title":"Calysto","year":2008,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":146,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Computer science; Scalability; Static analysis; Source lines of code; Automation; Static program analysis; Context (archaeology); Software; Software bug; Model checking; Code (set theory); Computer engineering; Programming language; Software development; Database; Set (abstract data type)","authors":[{"name":"Domagoj Babić","is_ca":true},{"name":"Alan J. Hu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03481987681072068,"gpt":0.236430062464799,"spread":0.2016101856540783,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001178145,0.001264159,0.0008444654,0.001505341,0.0008443953,0.002444507,0.003201999,0.001374948,0.04899959],"category_scores_gemma":[0.006284682,0.0008472028,0.0009143074,0.001038705,0.0009883365,0.003783002,0.002291144,0.002342733,0.01795174],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001248904,"about_ca_system_score_gemma":0.00208229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004978954,"about_ca_topic_score_gemma":0.007093893,"domain_scores_codex":[0.9982777,0.0001946433,0.00008530222,0.0003921557,0.0008404047,0.0002098404],"domain_scores_gemma":[0.9970528,0.000957906,0.0002754534,0.0007402353,0.0008053872,0.0001681635],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002075244,0.0002215998,0.004153631,0.001480059,0.0001301908,0.0005784253,0.0003975205,0.007527835,0.02897855,0.0515264,0.5722579,0.3306726],"study_design_scores_gemma":[0.0006778606,0.0003709186,0.003407751,0.0003010413,0.0001180292,0.001017398,0.0001218589,0.0872624,0.04723857,0.03809023,0.8212126,0.0001813753],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"software","genre_gemma":"other","genre_scores_codex":[0.02604278,0.002989466,0.2849006,0.002108302,0.00138068,0.0008472688,0.01994323,0.5140105,0.1477772],"genre_scores_gemma":[0.2749583,0.002287582,0.4131852,0.003646692,0.0004381402,0.001289468,0.08133406,0.08095688,0.1419037],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9510004,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1607747749","doi":"","title":"Specification-based regression test selection with risk analysis","year":2002,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":143,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Regression testing; Computer science; Test suite; Test Management Approach; Test case; Regression analysis; Traceability; Risk-based testing; Data mining; Unified Modeling Language; Reliability engineering; Machine learning; Software engineering; Programming language; Software; Engineering; Software development","authors":[{"name":"Yanping Chen","is_ca":true},{"name":"Robert L. Probert","is_ca":true},{"name":"D.Paul Sims","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01913616712242833,"gpt":0.228583529516784,"spread":0.2094473623943557,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008674227,0.00146969,0.0009218745,0.002945475,0.0004277234,0.001437449,0.001781068,0.001029746,0.002895568],"category_scores_gemma":[0.04424792,0.0005363142,0.001505372,0.001634698,0.0007565067,0.001752031,0.001474525,0.001362709,0.000984156],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007737642,"about_ca_system_score_gemma":0.001659028,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00163613,"about_ca_topic_score_gemma":0.001233578,"domain_scores_codex":[0.9809843,0.009761924,0.001102192,0.001080691,0.006491285,0.0005796758],"domain_scores_gemma":[0.9617823,0.02443567,0.00271347,0.004504957,0.006249919,0.0003136586],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0005602433,0.0006289987,0.02240417,0.0003609037,0.0003327901,0.0009759027,0.0004854088,0.2799241,0.02508632,0.04970523,0.005588543,0.6139475],"study_design_scores_gemma":[0.00008255549,0.0002413038,0.001503657,0.00003031638,0.00007853804,0.0004113345,0.00004956295,0.9673612,0.01464114,0.01348964,0.002076237,0.0000344787],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.009535216,0.00003885368,0.9874514,0.00009063936,0.000009381184,0.0001397163,0.00003787582,0.001729045,0.0009678466],"genre_scores_gemma":[0.2494701,0.0000677264,0.7478208,0.0001294481,0.00003125251,0.0004690629,0.0004226109,0.0004554355,0.001133592],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.008674227,"threshold_uncertainty_score":0.04587418,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1483143768","doi":"10.1007/3-540-45441-1_15","title":"A UML-Based Approach to System Testing","year":2001,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":137,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Computer science; Testability; Unified Modeling Language; Sequence diagram; Software engineering; Class diagram; Test case; Model-based testing; Programming language; Reliability engineering; Software; Engineering","authors":[{"name":"Lionel Briand","is_ca":true},{"name":"Yvan Labiche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0394705976459676,"gpt":0.2527500202852448,"spread":0.2132794226392772,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004117427,0.001348652,0.000883697,0.002809775,0.0009512659,0.003478307,0.003316424,0.00188939,0.008027661],"category_scores_gemma":[0.01263902,0.001257608,0.001458364,0.00175779,0.002666986,0.005536107,0.002569625,0.00387436,0.002493924],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001215664,"about_ca_system_score_gemma":0.001251017,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002102212,"about_ca_topic_score_gemma":0.003230898,"domain_scores_codex":[0.9947648,0.00209299,0.0003518382,0.0003442833,0.002273675,0.0001724074],"domain_scores_gemma":[0.9940951,0.003554156,0.0002056136,0.001147503,0.0008683754,0.0001292056],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004489854,0.0001731334,0.0003937495,0.000354592,0.0000561596,0.0002361618,0.000710111,0.02049403,0.00553875,0.6351124,0.01013761,0.3267484],"study_design_scores_gemma":[0.00006136856,0.00008805853,0.0003564983,0.0004312368,0.0001045465,0.0007450612,0.0001391803,0.1850387,0.008613855,0.7003487,0.1040185,0.00005430445],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0006674562,0.0003373709,0.9889812,0.0003660391,0.00006207315,0.0000676542,0.00002672617,0.0008511784,0.008640277],"genre_scores_gemma":[0.03844132,0.0008184659,0.9508404,0.0003399522,0.0001134583,0.0002266311,0.0001636548,0.0004577072,0.008598453],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.008027661,"threshold_uncertainty_score":0.02685523,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2102418068","doi":"10.1109/tse.2002.1049402","title":"Timed Wp-method: testing real-time systems","year":2002,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":135,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Concordia University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Nondeterministic algorithm; Correctness; Timed automaton; Automaton; System under test; Finite-state machine; Fault coverage; Real-time operating system; Test case; Real-time computing; Distributed computing; Algorithm; Theoretical computer science; Embedded system","authors":[{"name":"Abdeslam En‐Nouaary","is_ca":true},{"name":"Rachida Dssouli","is_ca":true},{"name":"Ferhat Khendek","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02672061398918014,"gpt":0.2342479938867207,"spread":0.2075273798975405,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001131038,0.001217668,0.0004707033,0.0007330701,0.0001707812,0.0004943058,0.002182449,0.0007573903,0.003019999],"category_scores_gemma":[0.00612649,0.0003305568,0.0007234346,0.0006353299,0.0007638785,0.001047783,0.000606149,0.0006283829,0.0004527835],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003880461,"about_ca_system_score_gemma":0.0008638132,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001372299,"about_ca_topic_score_gemma":0.0008773935,"domain_scores_codex":[0.9978585,0.0007516762,0.0001701859,0.0004434545,0.000670169,0.0001059785],"domain_scores_gemma":[0.9958978,0.002539444,0.0003507762,0.0007698603,0.0003586235,0.00008337678],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0007433625,0.0003630825,0.01002849,0.002164293,0.0003304199,0.001760901,0.0005391372,0.3207085,0.1341452,0.03921883,0.004840839,0.485157],"study_design_scores_gemma":[0.0001870222,0.0004940294,0.001952158,0.0001091016,0.0001038662,0.00119297,0.00009255286,0.8449922,0.1182679,0.02405252,0.008509327,0.00004621792],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01778726,0.00007905719,0.9782076,0.000040284,0.00002245992,0.0001611013,0.0001780987,0.002632268,0.0008918155],"genre_scores_gemma":[0.3384065,0.0001372041,0.6577503,0.00007711513,0.00002479691,0.0005239425,0.0007294917,0.000625727,0.001724841],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.003019999,"threshold_uncertainty_score":0.01010287,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2163289685","doi":"10.1109/icsm.2002.1167775","title":"Automating impact analysis and regression test selection based on UML designs","year":2003,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":131,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Unified Modeling Language; Regression testing; Selection (genetic algorithm); Test (biology); Regression analysis; Data mining; Programming language; Software engineering; Artificial intelligence; Machine learning; Software; Software development","authors":[{"name":"Lionel Briand","is_ca":true},{"name":"Yvan Labiche","is_ca":true},{"name":"G. Soccar","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02588388994970477,"gpt":0.3075896533459878,"spread":0.2817057633962831,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006420861,0.001277086,0.0009109926,0.004381758,0.0003946663,0.001199633,0.001229035,0.000945134,0.001387878],"category_scores_gemma":[0.03311447,0.0004371345,0.0007739286,0.001407224,0.0005836777,0.001105071,0.0008665691,0.0008373185,0.0004934518],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006673203,"about_ca_system_score_gemma":0.001370048,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001636165,"about_ca_topic_score_gemma":0.001879217,"domain_scores_codex":[0.9900649,0.005032775,0.0005483393,0.0006850033,0.003251315,0.0004176613],"domain_scores_gemma":[0.9587304,0.02749916,0.0048141,0.004260981,0.004267145,0.0004282784],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003155511,0.0005126572,0.02567015,0.0003699052,0.0001646706,0.0006148352,0.0006499781,0.05596564,0.05439333,0.006622933,0.001611517,0.8531089],"study_design_scores_gemma":[0.0002207185,0.0008648771,0.01651979,0.0001649236,0.0002540757,0.001012089,0.0002234903,0.8691687,0.08576689,0.0179903,0.007713232,0.000100987],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09070243,0.0002322759,0.8994449,0.0001667967,0.00002516777,0.000349474,0.00009385286,0.007187383,0.001797703],"genre_scores_gemma":[0.3857545,0.0001202507,0.6123359,0.00008640319,0.00002905635,0.0003731452,0.0002954872,0.0003216264,0.0006836753],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006420861,"threshold_uncertainty_score":0.03395718,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2741328617","doi":"10.1145/3106237.3106274","title":"Better test cases for better automated program repair","year":2017,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":126,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Overfitting; Metric (unit); Artificial intelligence; Benchmark (surveying); Test (biology); Machine learning; Software bug; Pattern recognition (psychology); Data mining; Software; Programming language; Engineering","authors":[{"name":"Jinqiu Yang","is_ca":true},{"name":"Alexey Zhikhartsev","is_ca":true},{"name":"Yuefei Liu","is_ca":true},{"name":"Lin Tan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04420587928543526,"gpt":0.3430140819929544,"spread":0.2988082027075191,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008221186,0.002627213,0.001238332,0.0061995,0.0004258998,0.002684202,0.002632215,0.001915935,0.006680829],"category_scores_gemma":[0.06377773,0.001028682,0.001389699,0.002004561,0.001150823,0.004943328,0.002040846,0.002168698,0.002271751],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001094512,"about_ca_system_score_gemma":0.001865573,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002871735,"about_ca_topic_score_gemma":0.00334958,"domain_scores_codex":[0.9854398,0.005918598,0.001402801,0.001656502,0.004851455,0.0007308448],"domain_scores_gemma":[0.9214358,0.04815075,0.007430682,0.01412758,0.007956191,0.0008989415],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008067993,0.00191927,0.05632962,0.001347721,0.0002367865,0.002689521,0.0008235904,0.1232042,0.08896373,0.01665823,0.02515726,0.6818634],"study_design_scores_gemma":[0.0005367082,0.001064177,0.0140745,0.0007117449,0.0002586574,0.002201181,0.0004753996,0.8056418,0.11735,0.01895848,0.03854674,0.0001805234],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1992685,0.001595839,0.7417679,0.001945663,0.0002224918,0.0007211114,0.001454006,0.04763563,0.005388782],"genre_scores_gemma":[0.466347,0.0004913442,0.5233479,0.0008011628,0.00009885047,0.0003150342,0.004065966,0.003135887,0.001396974],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.008221186,"threshold_uncertainty_score":0.04347825,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1967187747","doi":"10.1007/s10515-011-0093-0","title":"Prioritizing test cases with string distances","year":2011,"lang":"en","type":"article","venue":"Automated Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":125,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Computer Research Institute of Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Test suite; String (physics); Computer science; Random testing; Code (set theory); Test case; Algorithm; Property (philosophy); Suite; Test (biology); Data mining; Mathematics; Machine learning; Programming language","authors":[{"name":"Yves Ledru","is_ca":false},{"name":"Alexandre Petrenko","is_ca":true},{"name":"Sergiy Boroday","is_ca":true},{"name":"Nadine Mandran","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01989486036119929,"gpt":0.2132109271948698,"spread":0.1933160668336705,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004832741,0.002283864,0.002147098,0.007902329,0.0006572924,0.002162747,0.002861511,0.001696307,0.005037046],"category_scores_gemma":[0.04158924,0.0007593984,0.001316186,0.003534724,0.001036982,0.003661551,0.00255093,0.001575755,0.0009981827],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001219757,"about_ca_system_score_gemma":0.002483366,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002420063,"about_ca_topic_score_gemma":0.00447158,"domain_scores_codex":[0.9866136,0.00466725,0.001019888,0.001333214,0.005587504,0.0007785946],"domain_scores_gemma":[0.9504625,0.03825223,0.0024464,0.002701787,0.004791823,0.001345255],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003220589,0.0009394062,0.02023135,0.001252972,0.0003200633,0.001342836,0.0003347789,0.1143297,0.05425624,0.02603723,0.005791992,0.7719429],"study_design_scores_gemma":[0.0003120395,0.001182329,0.004474725,0.0001647068,0.0003097588,0.0009413636,0.0002659297,0.86316,0.059774,0.06433971,0.004987905,0.00008749074],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1964511,0.0008827432,0.7877273,0.0005868551,0.0001978494,0.0005529489,0.000527774,0.007391446,0.005681998],"genre_scores_gemma":[0.509549,0.0002315952,0.4851899,0.0002543803,0.0001018109,0.0002350657,0.001200708,0.0007752758,0.002462283],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007902329,"threshold_uncertainty_score":0.02555829,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2157056800","doi":"10.1145/2535838.2535857","title":"Symbolic optimization with SMT solvers","year":2014,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":122,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science; Satisfiability modulo theories; Programming language; Symbolic execution; Satisfiability; Relation (database); Theoretical computer science; Software; Subtyping; Concolic testing; Data mining","authors":[{"name":"Yi Li","is_ca":true},{"name":"Aws Albarghouthi","is_ca":true},{"name":"Zachary Kincaid","is_ca":true},{"name":"Arie Gurfinkel","is_ca":false},{"name":"Marsha Chećhik","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.005850794322010276,"gpt":0.1925467586558487,"spread":0.1866959643338385,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00127021,0.001278333,0.001075582,0.001170159,0.000500219,0.001541605,0.001370656,0.001174284,0.01023524],"category_scores_gemma":[0.006300593,0.0006491294,0.001547523,0.001575615,0.001600704,0.001380855,0.001959764,0.002155975,0.002478415],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001125077,"about_ca_system_score_gemma":0.001630685,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002354119,"about_ca_topic_score_gemma":0.004053712,"domain_scores_codex":[0.9983566,0.0006216952,0.00009919986,0.0001642696,0.0006202771,0.00013778],"domain_scores_gemma":[0.9975721,0.001759772,0.0001677698,0.0002702679,0.0001846919,0.00004535479],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001276743,0.00007733975,0.0005501937,0.0006334167,0.0001064415,0.0001857332,0.000117141,0.6318913,0.00319526,0.2547943,0.006218108,0.1021031],"study_design_scores_gemma":[0.00005249917,0.00002438641,0.00005589962,0.0000609123,0.00002595855,0.00004868919,0.00002635507,0.8480274,0.00229696,0.1378969,0.01147258,0.00001147048],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.004680654,0.0005557073,0.9678385,0.0005093544,0.0001114276,0.0000815178,0.0002346892,0.002282097,0.02370598],"genre_scores_gemma":[0.2055279,0.001341532,0.7801227,0.000357445,0.0002021979,0.0005148092,0.000827449,0.001032764,0.01007312],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01023524,"threshold_uncertainty_score":0.03424031,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3196239222","doi":"10.1145/3468264.3468591","title":"A comprehensive study of deep learning compiler bugs","year":2021,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":112,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"National Key Research and Development Program of China; National Natural Science Foundation of China","keywords":"Compiler; Computer science; Programming language; Boosting (machine learning); Deep learning; Context (archaeology); Code (set theory); Code generation; Parallel computing; Artificial intelligence; Operating system","authors":[{"name":"Qingchao Shen","is_ca":false},{"name":"Haoyang Ma","is_ca":false},{"name":"Junjie Chen","is_ca":false},{"name":"Yongqiang Tian","is_ca":true},{"name":"Shing-Chi Cheung","is_ca":false},{"name":"Xiang Chen","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03173090442745961,"gpt":0.283983725789503,"spread":0.2522528213620434,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003163037,0.0009864264,0.0009047389,0.001823818,0.0006628768,0.00155269,0.001748697,0.001501509,0.001691998],"category_scores_gemma":[0.02756425,0.0009137109,0.0007732537,0.001420534,0.002671411,0.004236799,0.001535836,0.003144266,0.0002856702],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00195161,"about_ca_system_score_gemma":0.002321689,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004740118,"about_ca_topic_score_gemma":0.003711327,"domain_scores_codex":[0.9977367,0.0004666333,0.0001396764,0.0003975914,0.001021998,0.0002373493],"domain_scores_gemma":[0.9842939,0.01028648,0.001690864,0.001361885,0.00207438,0.0002924992],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.0002006549,0.0003042237,0.02158268,0.001679014,0.000271313,0.0006008615,0.0006300489,0.334504,0.005439697,0.28861,0.01619087,0.3299867],"study_design_scores_gemma":[0.00002551133,0.0001234547,0.003662958,0.0003944419,0.00007936738,0.0004250145,0.0001112877,0.7019027,0.005616136,0.2769485,0.01064884,0.00006191315],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2516542,0.03492869,0.6749967,0.01395905,0.0005903266,0.0001080604,0.0005866203,0.003076228,0.02010017],"genre_scores_gemma":[0.8845194,0.008845049,0.09753366,0.001386161,0.0004888463,0.00009539775,0.0004958673,0.0007039179,0.005931674],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.996837,"threshold_uncertainty_score":0.01672792,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2162032875","doi":"10.1109/real.1998.739748","title":"Timed test cases generation based on state characterization technique","year":2002,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":112,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Concordia University; Université de Montréal","funders":"","keywords":"Computer science; Nondeterministic algorithm; Correctness; Test suite; Automaton; Timed automaton; Finite-state machine; Set (abstract data type); System under test; State (computer science); Granularity; Algorithm; Test case; Programming language; Theoretical computer science","authors":[{"name":"Abdeslam En‐Nouaary","is_ca":true},{"name":"Rachida Dssouli","is_ca":true},{"name":"Ferhat Khendek","is_ca":true},{"name":"A. Elqortobi","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0420348950945121,"gpt":0.2411333431361208,"spread":0.1990984480416087,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009736064,0.0007296949,0.0004808582,0.001334906,0.0002646327,0.0006788427,0.0009432107,0.000676952,0.002446793],"category_scores_gemma":[0.007567251,0.0003515655,0.0007767461,0.000679184,0.0006877309,0.0008950534,0.0006964862,0.0007471022,0.0004667101],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005466677,"about_ca_system_score_gemma":0.0007402841,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00125826,"about_ca_topic_score_gemma":0.0008673953,"domain_scores_codex":[0.9978988,0.0006284135,0.0002273236,0.0003589755,0.0007370344,0.0001494228],"domain_scores_gemma":[0.9946243,0.00319146,0.000515921,0.0007930617,0.0007826265,0.00009252421],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006571556,0.0004845457,0.007110064,0.0004606148,0.0001160222,0.001921044,0.0008402254,0.2627881,0.1468424,0.08649131,0.005022613,0.4872659],"study_design_scores_gemma":[0.0001281575,0.0002253841,0.0008020733,0.00004770215,0.00006615491,0.0006566898,0.00005618066,0.8705558,0.1002028,0.01993853,0.007271823,0.00004869655],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02302865,0.00002538953,0.9731506,0.0000496416,0.00001648292,0.0001929333,0.0001259878,0.002378957,0.00103128],"genre_scores_gemma":[0.3917471,0.00007184843,0.6049184,0.0000717466,0.00001667208,0.0006129507,0.0008972094,0.0004378239,0.001226189],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.002446793,"threshold_uncertainty_score":0.008185387,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2151635819","doi":"10.1145/1276958.1277172","title":"Automatic mutation test input data generation via ant colony","year":2007,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":109,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Polytechnique Montréal","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Computer science; Ant colony optimization algorithms; Mutation; Context (archaeology); Genetic algorithm; Java; Code coverage; Test data; Data mining; Machine learning; Artificial intelligence; Software","authors":[{"name":"Kamel Ayari","is_ca":true},{"name":"Salah Bouktif","is_ca":true},{"name":"Giuliano Antoniol","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05954161236665662,"gpt":0.3142915271483908,"spread":0.2547499147817342,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008070694,0.0005965366,0.0007308843,0.0007318659,0.0002891562,0.0004728952,0.001225664,0.0006879561,0.001019405],"category_scores_gemma":[0.003674129,0.000261608,0.0003643015,0.0004694014,0.0004163484,0.000625209,0.0007813065,0.0006384706,0.0002006032],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004122645,"about_ca_system_score_gemma":0.0004822922,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001765001,"about_ca_topic_score_gemma":0.001215603,"domain_scores_codex":[0.999011,0.0002885093,0.00005761526,0.0001254741,0.0004375211,0.00007982382],"domain_scores_gemma":[0.9979468,0.001061014,0.0002178367,0.0003653217,0.0003395186,0.00006952957],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002158057,0.0003093188,0.003520261,0.0001187705,0.00007348556,0.000391242,0.0001886411,0.4395267,0.09707081,0.007263181,0.001796388,0.4495254],"study_design_scores_gemma":[0.00002852139,0.00005150843,0.0003813149,0.000002995295,0.000009266332,0.00008736338,0.000008632305,0.9828398,0.01473152,0.00106244,0.0007880797,0.000008621438],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1203922,0.0001588556,0.872851,0.0001868814,0.00003517184,0.0001722671,0.00003829595,0.004112982,0.00205245],"genre_scores_gemma":[0.6901891,0.00008044231,0.3081494,0.00006967866,0.00001225008,0.0001813454,0.00009799333,0.0001457161,0.00107414],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.001765001,"threshold_uncertainty_score":0.004268289,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2036896594","doi":"10.1016/j.infsof.2013.02.006","title":"A systematic mapping study of web application testing","year":2013,"lang":"en","type":"article","venue":"Information and Software Technology","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":108,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; University of Calgary","funders":"","keywords":"Bibliometrics; Computer science; Data science; Field (mathematics); Schema (genetic algorithms); Information retrieval; Systematic review; World Wide Web; MEDLINE","authors":[{"name":"Vahid Garousi","is_ca":true},{"name":"Ali Mesbah","is_ca":true},{"name":"Aysu Betin-Can","is_ca":false},{"name":"Shabnam Mirshokraie","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01289810841019044,"gpt":0.2246790883964861,"spread":0.2117809799862957,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01488676,0.0005885662,0.0007608939,0.01382311,0.002120164,0.001803478,0.001478142,0.0008354512,0.002563067],"category_scores_gemma":[0.1481595,0.000689236,0.0008829731,0.009945185,0.001580298,0.003944129,0.003214261,0.0009369092,0.0004027855],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002516784,"about_ca_system_score_gemma":0.01299001,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01283384,"about_ca_topic_score_gemma":0.02385313,"domain_scores_codex":[0.9773312,0.01549255,0.00228693,0.001354751,0.003020723,0.0005138636],"domain_scores_gemma":[0.7169875,0.208155,0.01580119,0.02504527,0.0326645,0.001346576],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0006470575,0.002142526,0.2573214,0.009764352,0.0006498025,0.001301316,0.05774216,0.002243274,0.006384335,0.01335605,0.002627009,0.6458206],"study_design_scores_gemma":[0.0007399421,0.007726576,0.6413473,0.02284451,0.005513662,0.008370626,0.1083763,0.03678373,0.04393917,0.04479004,0.07913979,0.0004283645],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"review","genre_scores_codex":[0.8946654,0.006708629,0.08040228,0.0005532551,0.00007359427,0.003449189,0.001448755,0.0004524231,0.0122465],"genre_scores_gemma":[0.9499515,0.001430905,0.04537695,0.0001773526,0.00001122146,0.001021573,0.0006863892,0.00007565082,0.001268395],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9851133,"threshold_uncertainty_score":0.07872969,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2043839920","doi":"10.1016/j.jss.2014.01.010","title":"Web application testing: A systematic literature review","year":2014,"lang":"en","type":"article","venue":"Journal of Systems and Software","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":108,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Computer science; Web testing; Dependability; Systematic review; Field (mathematics); Web application; World Wide Web; Empirical research; Data science; Set (abstract data type); Web engineering; Software engineering; Information retrieval; Web application security; The Internet; Web development","authors":[{"name":"Serdar Doğan","is_ca":false},{"name":"Aysu Betin-Can","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01622432630285369,"gpt":0.2516204591044873,"spread":0.2353961328016336,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03115757,0.001972447,0.009482507,0.02916416,0.001351455,0.003711734,0.002842752,0.00296511,0.0040637],"category_scores_gemma":[0.1022371,0.001748256,0.007313124,0.01958571,0.001832617,0.00589163,0.00364054,0.00171132,0.000435062],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006112868,"about_ca_system_score_gemma":0.02830273,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008633388,"about_ca_topic_score_gemma":0.04275601,"domain_scores_codex":[0.9722,0.008684373,0.01228868,0.001688014,0.004520692,0.0006182463],"domain_scores_gemma":[0.8744955,0.08927678,0.0206444,0.002338519,0.01182777,0.001416928],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0002664805,0.00006723074,0.00205085,0.9208671,0.006822785,0.000270165,0.0007587859,0.000114569,0.0003378799,0.0002194546,0.002530351,0.06569418],"study_design_scores_gemma":[0.0003472883,0.0002503567,0.004463648,0.9331358,0.04676914,0.0005912998,0.001070074,0.0001066144,0.0002484458,0.0003277756,0.0126285,0.00006104765],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.003659199,0.992063,0.0006897793,0.000659054,0.0001696964,0.001591171,0.0007222653,0.00002023494,0.0004256118],"genre_scores_gemma":[0.03874077,0.9529697,0.003952641,0.001432231,0.0001070047,0.002029421,0.0005408912,0.00002152856,0.0002058408],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.03115757,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4254309703","doi":"10.1007/s10270-002-8208-5","title":"A UML-Based Approach to System Testing","year":2002,"lang":"en","type":"article","venue":"Software & Systems Modeling","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":107,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Computer science; Unified Modeling Language; Sequence diagram; Class diagram; Object Constraint Language; Software engineering; Applications of UML; Testability; Model-based testing; Context (archaeology); Test case; Programming language; UML tool; Test Management Approach; System testing; Activity diagram; Software development; Reliability engineering; Software; Software construction","authors":[{"name":"Lionel Briand","is_ca":true},{"name":"Yvan Labiche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09020794658056883,"gpt":0.2392422866175647,"spread":0.1490343400369959,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005754771,0.001140987,0.0008314478,0.003237966,0.0010371,0.00343138,0.003284051,0.001997009,0.005385592],"category_scores_gemma":[0.01883765,0.001087818,0.001647562,0.001489386,0.002281447,0.005301193,0.002524943,0.003775695,0.001603697],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001385985,"about_ca_system_score_gemma":0.001959346,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003171894,"about_ca_topic_score_gemma":0.00464994,"domain_scores_codex":[0.9917076,0.003501742,0.0005244516,0.0004615456,0.00353784,0.0002667792],"domain_scores_gemma":[0.9908714,0.004882849,0.0003701552,0.002089709,0.001560879,0.0002250476],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00007865132,0.0003291239,0.001054004,0.000337144,0.0001016066,0.000330659,0.0007536746,0.04209584,0.009260534,0.6227638,0.007445042,0.31545],"study_design_scores_gemma":[0.00008856354,0.0001477966,0.0006707059,0.0004470727,0.0001692706,0.000826291,0.0001557766,0.4362407,0.01347579,0.4791965,0.06850389,0.00007772334],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0008865838,0.000141517,0.9941871,0.0003244034,0.00004083728,0.00006556288,0.00002652638,0.0009066582,0.003420738],"genre_scores_gemma":[0.05121123,0.0003507326,0.9442317,0.0002958536,0.00007275666,0.0001947028,0.0001500489,0.0003355113,0.003157462],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005754771,"threshold_uncertainty_score":0.03043449,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2550120381","doi":"10.48550/arxiv.1611.01855","title":"Neuro-Symbolic Program Synthesis","year":2016,"lang":"en","type":"article","venue":"arXiv (Cornell University)","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":105,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Correctness; Computer science; Construct (python library); Program synthesis; Set (abstract data type); Representation (politics); Artificial neural network; Domain (mathematical analysis); Task (project management); String (physics); Theoretical computer science; Regular expression; Algorithm; Artificial intelligence; Programming language; Mathematics","authors":[{"name":"Emilio Parisotto","is_ca":true},{"name":"Abdelrahman Mohamed","is_ca":false},{"name":"Rishabh Singh","is_ca":false},{"name":"Lihong Li","is_ca":false},{"name":"Dengyong Zhou","is_ca":false},{"name":"Pushmeet Kohli","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06483697021880241,"gpt":0.1871020113852399,"spread":0.1222650411664375,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004038845,0.0005309367,0.0004102656,0.0004744075,0.0002366007,0.0005853231,0.000950753,0.0005739598,0.003561409],"category_scores_gemma":[0.001378647,0.0002587264,0.0006346236,0.0004157592,0.0008823032,0.0006789966,0.0006456896,0.0008045692,0.0005816882],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007548249,"about_ca_system_score_gemma":0.001054415,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001930987,"about_ca_topic_score_gemma":0.002886704,"domain_scores_codex":[0.9997224,0.00005412212,0.0000193764,0.00007253205,0.0001071558,0.00002429107],"domain_scores_gemma":[0.9995647,0.0002357416,0.00003373139,0.00008220335,0.00007155047,0.00001214449],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00009020478,0.00005375648,0.0005737974,0.0004174943,0.00005855246,0.0001384421,0.0001204616,0.6299821,0.01917082,0.1043234,0.00205753,0.2430135],"study_design_scores_gemma":[0.00001444719,0.00003114399,0.00008245699,0.00002612933,0.00001192691,0.00004369434,0.00001374507,0.9611396,0.006960609,0.02562074,0.006049076,0.000006581934],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01629951,0.0005864658,0.9708693,0.0002790758,0.00004749686,0.0000701491,0.0001435389,0.001803094,0.009901391],"genre_scores_gemma":[0.4104347,0.000990224,0.5809899,0.0001611474,0.00004000685,0.0003752187,0.0004606291,0.0002695195,0.006278722],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.003561409,"threshold_uncertainty_score":0.01191407,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2129718707","doi":"10.1016/j.infsof.2006.11.002","title":"A state-based approach to integration testing based on UML models","year":2006,"lang":"en","type":"article","venue":"Information and Software Technology","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":103,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Unified Modeling Language; Computer science; Integration testing; Test case; Class diagram; Reliability engineering; Sequence diagram; Data mining; Software; Software engineering; Programming language; Engineering; Machine learning","authors":[{"name":"Shaukat Ali","is_ca":false},{"name":"Lionel Briand","is_ca":true},{"name":"Muhammad Jaffar-Ur Rehman","is_ca":false},{"name":"Hajra Asghar","is_ca":false},{"name":"Muhammad Zohaib Iqbal","is_ca":false},{"name":"Aamer Nadeem","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01904630375427704,"gpt":0.2255805552280354,"spread":0.2065342514737583,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002857025,0.0009029382,0.0008676383,0.002057418,0.0008131822,0.002984883,0.002542443,0.001645657,0.003425379],"category_scores_gemma":[0.01006929,0.00111405,0.001611607,0.001102881,0.001896844,0.005173427,0.001951458,0.002385609,0.000660715],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001293743,"about_ca_system_score_gemma":0.0016788,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004627174,"about_ca_topic_score_gemma":0.00678359,"domain_scores_codex":[0.9963638,0.001310233,0.0002313645,0.0003521442,0.001571455,0.00017109],"domain_scores_gemma":[0.9930043,0.003755382,0.0003810296,0.001715597,0.001015552,0.000128221],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0002966505,0.0005033473,0.00230026,0.0002874185,0.0001976484,0.0003977524,0.00125007,0.2456209,0.01894991,0.4850573,0.00315136,0.2419875],"study_design_scores_gemma":[0.00003085292,0.00007602695,0.0002175268,0.00006790968,0.0001011058,0.000105459,0.00003963865,0.8896341,0.007501062,0.09858515,0.003602251,0.0000389855],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.003976123,0.00005480391,0.9924088,0.0001203234,0.00001693576,0.00005301403,0.00003239768,0.001656324,0.001681283],"genre_scores_gemma":[0.2903992,0.0002086547,0.705842,0.0001386617,0.00003282012,0.0002828412,0.0002051907,0.0003890171,0.002501593],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004627174,"threshold_uncertainty_score":0.0151096,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2145733341","doi":"10.1145/2786805.2786858","title":"Assertions are strongly correlated with test suite effectiveness","year":2015,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":101,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Test suite; Assertion; Test (biology); Computer science; Suite; Java; Test case; Code coverage; Code (set theory); Unit testing; Programming language; Algorithm; Machine learning; Software; Regression analysis","authors":[{"name":"Yucheng Zhang","is_ca":true},{"name":"Ali Mesbah","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02982942736725199,"gpt":0.2597681029273936,"spread":0.2299386755601416,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01888615,0.0008402126,0.0007817411,0.00498593,0.0003354822,0.001766364,0.0009264762,0.001070926,0.001474788],"category_scores_gemma":[0.2660828,0.0004806139,0.000817842,0.00261146,0.001097516,0.003293591,0.001256342,0.001214238,0.000467437],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006190499,"about_ca_system_score_gemma":0.0007083042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001099896,"about_ca_topic_score_gemma":0.001208226,"domain_scores_codex":[0.9673128,0.009785358,0.004322873,0.00276408,0.01450468,0.001310266],"domain_scores_gemma":[0.3411236,0.5635101,0.05368936,0.01333255,0.02408697,0.004257432],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006011669,0.0003185691,0.903013,0.0003789793,0.0005988287,0.0005203988,0.0007475669,0.01785992,0.008079708,0.00106271,0.00080109,0.06601793],"study_design_scores_gemma":[0.00005405649,0.001210828,0.898567,0.0001417689,0.0005870701,0.001667434,0.0006928974,0.07965475,0.01170067,0.003708768,0.001933971,0.00008071618],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9846379,0.0009310448,0.01091212,0.0002908989,0.00002133453,0.00004812701,0.0003092982,0.0004895542,0.002359699],"genre_scores_gemma":[0.9949485,0.0001460286,0.004023796,0.00003454202,0.00002790815,0.00003070834,0.0005308881,0.00008489111,0.0001727782],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01888615,"threshold_uncertainty_score":0.0998807,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1959797194","doi":"10.1007/978-3-642-31057-7_30","title":"Application-Only Call Graph Construction","year":2012,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":100,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"Call graph; Computer science; Graph; Theoretical computer science; Programming language","authors":[{"name":"Karim Ali","is_ca":true},{"name":"Ondřej Lhoták","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0139956444234305,"gpt":0.2447250528338945,"spread":0.230729408410464,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007954176,0.0009853579,0.0008607149,0.001564376,0.0009733699,0.001641983,0.002162635,0.00116704,0.02265947],"category_scores_gemma":[0.00418896,0.0009716447,0.001317771,0.001341829,0.00115905,0.002925879,0.003744458,0.00290718,0.01029675],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005116172,"about_ca_system_score_gemma":0.002039242,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001283436,"about_ca_topic_score_gemma":0.002490622,"domain_scores_codex":[0.9984871,0.0002385211,0.00007052082,0.0002981671,0.0006563514,0.0002493727],"domain_scores_gemma":[0.9966983,0.0006410283,0.00008469279,0.001904811,0.000542235,0.0001289464],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002545671,0.0003752061,0.001802908,0.0007564115,0.000108594,0.0005929249,0.0004748183,0.01082784,0.05243405,0.1917922,0.05558983,0.6849906],"study_design_scores_gemma":[0.0001034433,0.0002488744,0.002820885,0.0002531766,0.0003403493,0.002208442,0.0002858353,0.1787696,0.1516557,0.3693721,0.2937928,0.0001487637],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0102824,0.0001566943,0.9366947,0.0003023525,0.0002087081,0.0002476018,0.0004117626,0.02088309,0.03081264],"genre_scores_gemma":[0.2515215,0.0004012136,0.6940723,0.0006189496,0.0001333429,0.000377951,0.002680143,0.009512159,0.04068249],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02265947,"threshold_uncertainty_score":0.07580352,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4394015350","doi":"10.1016/j.infsof.2024.107468","title":"Effective test generation using pre-trained Large Language Models and mutation testing","year":2024,"lang":"en","type":"article","venue":"Information and Software Technology","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":97,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Polytechnique Montréal","funders":"Fonds de recherche du Québec; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs; Canadian Institute for Advanced Research","keywords":"Test (biology); Mutation; Computer science; Natural language processing; Genetics; Biology; Gene","authors":[{"name":"Arghavan Moradi Dakhel","is_ca":true},{"name":"Amin Nikanjam","is_ca":true},{"name":"Vahid Majdinasab","is_ca":true},{"name":"Foutse Khomh","is_ca":true},{"name":"Michel C. Desmarais","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01684748034253352,"gpt":0.2775876674362418,"spread":0.2607401870937083,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001237452,0.001579311,0.001016352,0.00133628,0.0003595494,0.0007329915,0.001884496,0.00151937,0.001727021],"category_scores_gemma":[0.008385114,0.0004827386,0.001178309,0.0006041561,0.0006916634,0.001281387,0.001002052,0.001509285,0.0008576497],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008024166,"about_ca_system_score_gemma":0.001382725,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004964372,"about_ca_topic_score_gemma":0.006390506,"domain_scores_codex":[0.9986789,0.0004145021,0.0000696754,0.0004246544,0.0002971177,0.0001151506],"domain_scores_gemma":[0.9941551,0.004301644,0.0003445081,0.0004481661,0.0005950739,0.0001554849],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00046,0.0006600637,0.008439554,0.0004469495,0.0001737007,0.001282466,0.000273793,0.3925345,0.03624671,0.0020628,0.008417867,0.5490016],"study_design_scores_gemma":[0.00003307246,0.00009789668,0.0005333508,0.00001355154,0.0000236734,0.0001596766,0.00002871656,0.9910425,0.005875171,0.001592714,0.0005876768,0.00001200032],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1823681,0.0009965552,0.7857834,0.000658948,0.0001264453,0.0003131207,0.000635537,0.02716559,0.001952246],"genre_scores_gemma":[0.7050749,0.0001941245,0.2882021,0.0007379483,0.00006047526,0.0003441855,0.002924585,0.0006687777,0.001792911],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.004964372,"threshold_uncertainty_score":0.009870946,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1522741019","doi":"10.1002/stvr.1572","title":"Coverage‐based regression test case selection, minimization and prioritization: a case study on an industrial system","year":2015,"lang":"en","type":"article","venue":"Software Testing Verification and Reliability","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":95,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; Fonds National de la Recherche Luxembourg","keywords":"Regression testing; Minification; Selection (genetic algorithm); Computer science; Prioritization; Fault detection and isolation; Suite; Reliability engineering; Test suite; Regression; Fault (geology); Regression analysis; Data mining; Machine learning; Test case; Artificial intelligence; Statistics; Engineering; Software; Mathematics; Software system","authors":[{"name":"Daniel Di Nardo","is_ca":false},{"name":"Nadia Alshahwan","is_ca":false},{"name":"Lionel Briand","is_ca":false},{"name":"Yvan Labiche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08433280682020698,"gpt":0.3118294633298048,"spread":0.2274966565095978,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00236282,0.0004769283,0.0003753844,0.001163186,0.0005224155,0.0004736378,0.0009475021,0.0007806709,0.001036378],"category_scores_gemma":[0.009054066,0.000261037,0.0004380928,0.0008853189,0.0005641717,0.0004912508,0.0003807077,0.0004981832,0.0001413628],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009127594,"about_ca_system_score_gemma":0.0007646199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007404279,"about_ca_topic_score_gemma":0.008795127,"domain_scores_codex":[0.9977835,0.001092304,0.0001004659,0.0001448867,0.0007050007,0.000173804],"domain_scores_gemma":[0.985876,0.01149689,0.0007148496,0.0006826772,0.001027153,0.0002023579],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001622368,0.002581368,0.05502882,0.0008289422,0.0002315673,0.01098338,0.001896136,0.5816472,0.06314454,0.00491456,0.002932246,0.2741889],"study_design_scores_gemma":[0.000512466,0.003862686,0.02980914,0.00007534023,0.0002227781,0.003629441,0.00106072,0.8720822,0.08189059,0.001645526,0.00513286,0.00007624548],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9730856,0.000136776,0.02413686,0.0001916237,0.00000695634,0.0001601305,0.00007457532,0.0002639738,0.001943458],"genre_scores_gemma":[0.9740158,0.00007177189,0.02511113,0.00002566575,0.000004507941,0.00005683756,0.00007512997,0.00002904397,0.0006099965],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007404279,"threshold_uncertainty_score":0.01472235,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2099441126","doi":"10.1109/tse.2003.1214327","title":"General test result checking with log file analysis","year":2003,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":95,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Computer science; Random testing; Formalism (music); Programming language; Unit testing; Operating system; Software; Test case","authors":[{"name":"James H. Andrews","is_ca":true},{"name":"Yingjun Zhang","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01114718760434826,"gpt":0.215001202784079,"spread":0.2038540151797307,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005819703,0.00161318,0.001298364,0.003277684,0.0007152322,0.003221233,0.004041425,0.00168409,0.005121144],"category_scores_gemma":[0.0368065,0.0008092988,0.001706109,0.00183454,0.003313489,0.006775377,0.003306326,0.00195161,0.001431096],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001318562,"about_ca_system_score_gemma":0.002989001,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002180503,"about_ca_topic_score_gemma":0.001605702,"domain_scores_codex":[0.9863593,0.004570359,0.001313333,0.001529735,0.005437585,0.0007896477],"domain_scores_gemma":[0.9638414,0.01705532,0.003144664,0.01255218,0.003080451,0.0003258988],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001872118,0.0009540087,0.01767866,0.00185562,0.0003008865,0.001802247,0.001051244,0.1117063,0.0617876,0.1840515,0.01417102,0.6027688],"study_design_scores_gemma":[0.0002562022,0.0006021031,0.002331463,0.0003287669,0.0001928569,0.002089937,0.0001342801,0.7014862,0.1226012,0.1508023,0.0189905,0.0001841779],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.007896787,0.00007660389,0.9808736,0.00009210307,0.00001678504,0.0001476252,0.0001269231,0.00981705,0.000952517],"genre_scores_gemma":[0.3169245,0.000163728,0.6777416,0.0002509195,0.00008238597,0.0006486533,0.0009724647,0.001433479,0.001782355],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005819703,"threshold_uncertainty_score":0.03077793,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2952903800","doi":"10.1109/icst.2019.00019","title":"BugsJS: a Benchmark of JavaScript Bugs","year":2019,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":94,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"JavaScript; Unobtrusive JavaScript; Computer science; Benchmark (surveying); Programming language; Test case; Software bug; Unit testing; Rich Internet application; Operating system; Database; Software; Machine learning","authors":[{"name":"Péter Gyimesi","is_ca":false},{"name":"Béla Vancsics","is_ca":false},{"name":"Andrea Stocco","is_ca":true},{"name":"Davood Mazinanian","is_ca":true},{"name":"Árpád Beszédes","is_ca":false},{"name":"Rudolf Ferenć","is_ca":false},{"name":"Ali Mesbah","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.004992930306730839,"gpt":0.2021300423044523,"spread":0.1971371119977214,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00566814,0.001664238,0.0006491041,0.005360328,0.0006792318,0.001141416,0.00290446,0.001272778,0.001066714],"category_scores_gemma":[0.03439905,0.0005437106,0.0009441368,0.004233945,0.001132881,0.001815648,0.00171177,0.001414875,0.0009295581],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007938062,"about_ca_system_score_gemma":0.001774942,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004403885,"about_ca_topic_score_gemma":0.006278822,"domain_scores_codex":[0.9892736,0.002322168,0.001513904,0.001497792,0.004823578,0.0005689814],"domain_scores_gemma":[0.967399,0.0149638,0.00405836,0.00504528,0.0072895,0.001244148],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002865691,0.003297372,0.1990172,0.01433499,0.001277747,0.00260231,0.003616604,0.0783501,0.0882781,0.01070513,0.1664803,0.4291745],"study_design_scores_gemma":[0.001067933,0.004410102,0.2517635,0.001921781,0.0008372967,0.005122528,0.002112906,0.3928814,0.1320781,0.01524384,0.1920065,0.0005542094],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"dataset","genre_scores_codex":[0.7875907,0.009624546,0.08969591,0.001048601,0.0006564929,0.00136284,0.03692102,0.05961674,0.01348318],"genre_scores_gemma":[0.7034864,0.002238874,0.1726481,0.0006038807,0.0001463137,0.00108743,0.1071049,0.008819089,0.003865],"genre_candidate":"dataset","genre_consensus":null,"teacher_disagreement_score":0.00566814,"threshold_uncertainty_score":0.02997631,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2535681473","doi":"10.1109/mutation.2006.10","title":"Mutation Operators for Concurrent Java (J2SE 5.0)","year":2006,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":94,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Computer science; Concurrency; Java; Programming language; Java concurrency; Operating system; Semaphore; Real time Java","authors":[{"name":"Jeremy S. Bradbury","is_ca":true},{"name":"James R. Cordy","is_ca":true},{"name":"Juergen Dingel","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0165711930763524,"gpt":0.2721297957116077,"spread":0.2555586026352553,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004192478,0.001001134,0.000583102,0.001575425,0.001002511,0.001618812,0.001593725,0.001076985,0.002556069],"category_scores_gemma":[0.01141091,0.000668151,0.001609945,0.0006652701,0.001544031,0.002076161,0.002327705,0.00204825,0.0006686154],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008347331,"about_ca_system_score_gemma":0.001624918,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001794365,"about_ca_topic_score_gemma":0.002275167,"domain_scores_codex":[0.9964269,0.0005533104,0.0004736522,0.0004160466,0.001873109,0.0002570158],"domain_scores_gemma":[0.9928798,0.003516955,0.00107181,0.0009537823,0.001343089,0.0002346707],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001013821,0.0007572261,0.01641675,0.001795237,0.0002692909,0.002823906,0.002669827,0.01876141,0.1493121,0.2631988,0.03575389,0.5072277],"study_design_scores_gemma":[0.0006978325,0.001352208,0.01030784,0.0006692333,0.0006099861,0.009678012,0.0005739609,0.2368045,0.2253244,0.215061,0.2982759,0.000645201],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0344491,0.0003672215,0.935554,0.0004217557,0.0003340103,0.000463982,0.0003285188,0.02279305,0.005288229],"genre_scores_gemma":[0.2338397,0.0004123037,0.7541844,0.0007362658,0.0001303168,0.0005803398,0.0005967118,0.005138699,0.004381282],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004192478,"threshold_uncertainty_score":0.02217215,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2156296121","doi":"10.1109/issre.2007.31","title":"Using Machine Learning to Support Debugging with Tarantula","year":2007,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":93,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Debugging; Computer science; Statement (logic); Ranking (information retrieval); Test (biology); Software bug; Decision tree; Fault (geology); Fault tree analysis; Machine learning; Algorithmic program debugging; Test case; Tree (set theory); Artificial intelligence; Programming language; Reliability engineering; Software; Engineering","authors":[{"name":"Lionel Briand","is_ca":false},{"name":"Yvan Labiche","is_ca":true},{"name":"Xuetao Liu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04028565877204957,"gpt":0.3031118466118101,"spread":0.2628261878397605,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002923959,0.001289028,0.0007673798,0.002519254,0.0006002262,0.001232724,0.001895138,0.0009719076,0.002011527],"category_scores_gemma":[0.01864794,0.0005734838,0.0009057835,0.001159567,0.0006054129,0.002532661,0.0009399613,0.001710722,0.0009168853],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006433317,"about_ca_system_score_gemma":0.001280003,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003391907,"about_ca_topic_score_gemma":0.005073758,"domain_scores_codex":[0.9981624,0.0007310463,0.0001553871,0.0003589726,0.0004771676,0.0001149956],"domain_scores_gemma":[0.9871875,0.008828173,0.001224531,0.00150875,0.001087885,0.0001632478],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004968736,0.0005237059,0.01059561,0.0003809916,0.0001821126,0.000607656,0.0005206378,0.330467,0.01555355,0.01876792,0.004564636,0.6173393],"study_design_scores_gemma":[0.00002187197,0.00007980824,0.0004993915,0.00003015911,0.0000226171,0.0001189941,0.00001794314,0.9801364,0.008709285,0.008789463,0.001550323,0.00002384158],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02118737,0.0000857083,0.9669476,0.0001962566,0.00002017898,0.00007932742,0.00009745089,0.01061578,0.000770311],"genre_scores_gemma":[0.2003576,0.00008406168,0.7982815,0.0001051557,0.00001968876,0.00008045883,0.0002851178,0.0002376564,0.0005488443],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003391907,"threshold_uncertainty_score":0.01546359,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2160197556","doi":"10.1109/tc.2006.80","title":"Optimizing the length of checking sequences","year":2006,"lang":"en","type":"article","venue":"IEEE Transactions on Computers","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":93,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"Engineering and Physical Sciences Research Council; Natural Sciences and Engineering Research Council of Canada; Leverhulme Trust","keywords":"Sequence (biology); State (computer science); Set (abstract data type); Algorithm; Finite-state machine; Mathematics; Computer science; Combinatorics","authors":[{"name":"Robert M. Hierons","is_ca":false},{"name":"Hasan Ural","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02136507433847591,"gpt":0.2432458928702451,"spread":0.2218808185317692,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001816112,0.001244268,0.0009740067,0.001257092,0.000626492,0.001231808,0.001626878,0.0009370499,0.003306209],"category_scores_gemma":[0.01124379,0.0007815522,0.0006866411,0.001228095,0.000739606,0.002092422,0.0009561805,0.001148615,0.0009924689],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001362737,"about_ca_system_score_gemma":0.002501872,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001687552,"about_ca_topic_score_gemma":0.003061296,"domain_scores_codex":[0.9974689,0.0006963806,0.0002697932,0.0006208033,0.0006153698,0.0003287031],"domain_scores_gemma":[0.9915406,0.004781183,0.001207061,0.0007517575,0.001331235,0.0003882332],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001706514,0.0005095878,0.008256962,0.0006332221,0.0001401059,0.0003555253,0.0003478143,0.6055328,0.07574732,0.01024674,0.003039558,0.2934838],"study_design_scores_gemma":[0.0001446665,0.0009740488,0.002371745,0.00008764779,0.0001511088,0.0003784322,0.0002414409,0.9167105,0.06069639,0.01290304,0.005295115,0.00004583855],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2878277,0.001356049,0.6954608,0.0005557786,0.0001356595,0.0005271719,0.0005282341,0.003715159,0.009893345],"genre_scores_gemma":[0.6097521,0.0004035428,0.3842092,0.0001502993,0.0000416667,0.0003018725,0.0008898994,0.0007792449,0.003472139],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003306209,"threshold_uncertainty_score":0.01106036,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1920048997","doi":"10.1109/ase.1998.732614","title":"Testing using log file analysis: tools, methods, and issues","year":2002,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":93,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada; University of British Columbia","keywords":"Computer science; Scope (computer science); Software; Software testing; Programming language; Database; Software engineering","authors":[{"name":"James H. Andrews","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1628634645043925,"gpt":0.373331725608945,"spread":0.2104682611045524,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04923257,0.003170616,0.002504949,0.009950952,0.001580804,0.01219941,0.009852931,0.006032825,0.003500575],"category_scores_gemma":[0.1392363,0.002742856,0.001438868,0.008230467,0.01558924,0.03295551,0.004542779,0.006633731,0.002555646],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00230167,"about_ca_system_score_gemma":0.003867541,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002818757,"about_ca_topic_score_gemma":0.00134419,"domain_scores_codex":[0.9475852,0.02944271,0.002884181,0.003026897,0.01621464,0.0008462576],"domain_scores_gemma":[0.7355471,0.2088661,0.00670854,0.03276064,0.01385947,0.002258152],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001487749,0.0004250815,0.007809171,0.001575075,0.0001050532,0.0003391027,0.001740396,0.00944681,0.002221385,0.349364,0.0100412,0.6167839],"study_design_scores_gemma":[0.000125898,0.0003087484,0.002675294,0.00212781,0.00008194529,0.002355851,0.001335058,0.1568738,0.008145825,0.7640285,0.06156414,0.0003771456],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002210114,0.005693673,0.9816868,0.005290472,0.0001186815,0.0001331348,0.00006933638,0.002608334,0.002189419],"genre_scores_gemma":[0.07682523,0.01003235,0.9072332,0.0008851908,0.0006124445,0.0009335654,0.000199501,0.001167946,0.002110439],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04923257,"threshold_uncertainty_score":0.2603698,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1880492488","doi":"10.1120/jacmp.v14i1.4052","title":"Automated IMRT planning with regional optimization using planning scripts","year":2013,"lang":"en","type":"article","venue":"Journal of Applied Clinical Medical Physics","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":92,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"London Health Sciences Centre; Western University","funders":"Cancer Care Ontario","keywords":"Scripting language; Radiation treatment planning; Computer science; Medical physics; Medicine; Radiation therapy; Radiology; Programming language","authors":[{"name":"Ilma Xhaferllari","is_ca":true},{"name":"Eugene Wong","is_ca":true},{"name":"K Bzdusek","is_ca":false},{"name":"Michael Lock","is_ca":true},{"name":"Jeff Z. Chen","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09085317414621082,"gpt":0.3731214052892368,"spread":0.282268231143026,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001402705,0.001801383,0.0006308408,0.0007554092,0.0004861247,0.001285399,0.001983393,0.0006246355,0.02228814],"category_scores_gemma":[0.003395978,0.0008943655,0.0007716178,0.0005465184,0.000445328,0.0005791225,0.0007245499,0.001151217,0.005489206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007132068,"about_ca_system_score_gemma":0.00148334,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00247431,"about_ca_topic_score_gemma":0.002199225,"domain_scores_codex":[0.9986652,0.0003363712,0.0001288578,0.0002805091,0.0004876069,0.0001014815],"domain_scores_gemma":[0.9978951,0.0009065635,0.0001853464,0.0003001989,0.000644133,0.00006854357],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0007314238,0.0002699255,0.001845581,0.0007569875,0.0001505432,0.0004475575,0.0005363591,0.3128284,0.05920926,0.01070829,0.04922342,0.5632923],"study_design_scores_gemma":[0.0001233734,0.0001326033,0.0006746339,0.0000564226,0.00003607928,0.0001646646,0.00003546468,0.9010256,0.05398038,0.003137185,0.04055973,0.00007387879],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00451134,0.00005418109,0.9392417,0.0000330975,0.00004472108,0.000445073,0.0007021916,0.0504964,0.004471375],"genre_scores_gemma":[0.06572435,0.0000758742,0.9169537,0.00009497952,0.00002381157,0.001046959,0.002003761,0.009411376,0.004665265],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02228814,"threshold_uncertainty_score":0.0745613,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2056488032","doi":"10.1145/1391984.1391985","title":"Unit-level test adequacy criteria for visual dataflow languages and a testing methodology","year":2008,"lang":"en","type":"article","venue":"ACM Transactions on Software Engineering and Methodology","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":92,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Dalhousie University","funders":"","keywords":"Computer science; Dataflow; Context (archaeology); Unit testing; Visual programming language; Empirical research; Software engineering; Programming language; Software","authors":[{"name":"Marcel Karam","is_ca":false},{"name":"Trevor J. Smedley","is_ca":true},{"name":"Sergiu M. Dascalu","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3232977272659157,"gpt":0.4163952044641103,"spread":0.09309747719819456,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01072826,0.0009186876,0.0007564116,0.006618063,0.0006563935,0.001936338,0.00150203,0.001514522,0.001749335],"category_scores_gemma":[0.08693962,0.0003997194,0.001073803,0.001817541,0.00213775,0.002496471,0.001566842,0.0009339405,0.0002209674],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001277839,"about_ca_system_score_gemma":0.001490789,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001815057,"about_ca_topic_score_gemma":0.001851716,"domain_scores_codex":[0.9824097,0.008039705,0.001972932,0.0009765016,0.006048907,0.0005523178],"domain_scores_gemma":[0.8961833,0.08041725,0.006704325,0.004858478,0.01072586,0.001110794],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001181941,0.0009115228,0.07097267,0.001148968,0.0002697564,0.001300904,0.0018657,0.2241284,0.07255206,0.143042,0.00284731,0.4797787],"study_design_scores_gemma":[0.0001184743,0.0008818285,0.007116665,0.0001974844,0.00008042854,0.001019636,0.0004082145,0.9051265,0.038119,0.04430988,0.002561794,0.00006020431],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0924143,0.0002220898,0.9037453,0.0001838755,0.0000123966,0.0002551624,0.0001053,0.0007693805,0.002292308],"genre_scores_gemma":[0.6517582,0.00005791319,0.3469537,0.00008101921,0.00002330335,0.00041348,0.0002457433,0.000115079,0.0003515071],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01072826,"threshold_uncertainty_score":0.05673712,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4378942602","doi":"10.1145/3597926.3598135","title":"How Effective Are Neural Networks for Fixing Security Vulnerabilities","year":2023,"lang":"en","type":"preprint","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":90,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; York University","funders":"National Science Foundation","keywords":"Computer science; Task (project management); Vulnerability (computing); Code (set theory); Software security assurance; Secure coding; Computer security; Source code; Software; Artificial neural network; Software bug; Automation; Programming language; Software engineering; Artificial intelligence; Information security; Engineering; Security service","authors":[{"name":"Yi Wu","is_ca":false},{"name":"Nan Jiang","is_ca":false},{"name":"Hung Viet Pham","is_ca":true},{"name":"Thibaud Lutellier","is_ca":true},{"name":"Jordan Davis","is_ca":false},{"name":"Lin Tan","is_ca":false},{"name":"Petr Babkin","is_ca":false},{"name":"Sameena Shah","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03806044719484671,"gpt":0.2861846731068167,"spread":0.24812422591197,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003086173,0.001406625,0.0007414048,0.001078236,0.000422634,0.002305769,0.0009718284,0.003234642,0.003046815],"category_scores_gemma":[0.02958437,0.0005349159,0.0004321634,0.0005418243,0.00108227,0.008299098,0.000807499,0.002060202,0.001347335],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009698254,"about_ca_system_score_gemma":0.001147727,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005457677,"about_ca_topic_score_gemma":0.006748009,"domain_scores_codex":[0.9981365,0.0005960696,0.0000737486,0.0005323411,0.0003951627,0.0002661184],"domain_scores_gemma":[0.991687,0.005540934,0.0006867127,0.0009766413,0.0007986865,0.0003100418],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001035269,0.0005142996,0.021983,0.0005151174,0.0005800984,0.0001390658,0.0001827981,0.2689991,0.01279367,0.01418246,0.02180167,0.6572735],"study_design_scores_gemma":[0.00008922849,0.0003129583,0.007123369,0.0001959837,0.0002423819,0.0001637897,0.0002255493,0.9087669,0.0148015,0.06278399,0.00522646,0.00006792458],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5140172,0.02228146,0.3751123,0.0331829,0.001755323,0.0001646163,0.001619797,0.007073826,0.0447926],"genre_scores_gemma":[0.9543778,0.002882225,0.03738718,0.0009998658,0.0002389659,0.00004452619,0.0004757416,0.0002639316,0.003329819],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005457677,"threshold_uncertainty_score":0.01632148,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2067617772","doi":"10.1109/tse.2014.2363479","title":"Instance Generator and Problem Representation to Improve Object Oriented Code Coverage","year":2014,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":86,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Computer science; Unit testing; Test case; Programming language; Code generation; Java; Code coverage; Object-oriented programming; Data structure; Source code; Generator (circuit theory); Software; Theoretical computer science; Machine learning; Operating system","authors":[{"name":"Abdelilah Sakti","is_ca":true},{"name":"Gilles Pesant","is_ca":true},{"name":"Yann‐Gaël Guéhéneuc","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.009565537506992744,"gpt":0.2305431465418212,"spread":0.2209776090348285,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002328164,0.001017576,0.000831974,0.001531197,0.0003317904,0.001015726,0.001676112,0.001218837,0.003438442],"category_scores_gemma":[0.01602101,0.0004704888,0.001198612,0.001181567,0.0008833914,0.0014142,0.001639137,0.00128098,0.0005158788],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007584185,"about_ca_system_score_gemma":0.001387478,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001460768,"about_ca_topic_score_gemma":0.0017441,"domain_scores_codex":[0.9980094,0.000918098,0.0001307244,0.0002694142,0.0005261975,0.0001461344],"domain_scores_gemma":[0.9907134,0.007262501,0.0003667025,0.0008141357,0.0007032736,0.0001399813],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003591938,0.0005333858,0.006147843,0.0005472008,0.0001105115,0.0004538224,0.0004198254,0.6321445,0.01255531,0.03774348,0.008172878,0.3008122],"study_design_scores_gemma":[0.00007881963,0.00006069355,0.0002468131,0.00002027515,0.00002598073,0.00008863545,0.00003218246,0.9833027,0.004014054,0.009934347,0.002187581,0.000007931818],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.06145282,0.0002558603,0.9303147,0.0004104004,0.00004355583,0.0003015467,0.0002720099,0.00420569,0.002743379],"genre_scores_gemma":[0.3087514,0.0001466668,0.6867939,0.0001907984,0.00004021052,0.000581838,0.001132632,0.0007125461,0.001649964],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.003438442,"threshold_uncertainty_score":0.01231259,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2121504314","doi":"10.1145/2642937.2642991","title":"Leveraging existing tests in automated test generation for web applications","year":2014,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":82,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung","keywords":"Test suite; Computer science; Web crawler; Crawling; Code coverage; Test (biology); Web testing; Test harness; Random testing; Test Management Approach; Test case; Domain (mathematical analysis); Generator (circuit theory); Software engineering; Event (particle physics); Data mining; Machine learning; Web page; Programming language; World Wide Web; Software; Power (physics); Web development; Software development; Web application security","authors":[{"name":"Amin Milani Fard","is_ca":true},{"name":"Mehdi Mirzaaghaei","is_ca":true},{"name":"Ali Mesbah","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0607930062329493,"gpt":0.3199788099738178,"spread":0.2591858037408685,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007126146,0.001631827,0.001053127,0.005579936,0.00060581,0.001752936,0.002421212,0.001447701,0.001926212],"category_scores_gemma":[0.05351976,0.000873188,0.001119379,0.002355138,0.001529687,0.002698276,0.001530791,0.00135616,0.0008156699],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009592399,"about_ca_system_score_gemma":0.001858906,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002806534,"about_ca_topic_score_gemma":0.004962389,"domain_scores_codex":[0.9882414,0.006072228,0.0008140591,0.001320131,0.003089359,0.0004628009],"domain_scores_gemma":[0.9189354,0.05832765,0.004941287,0.0112599,0.00589492,0.0006409101],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000459474,0.0009827035,0.02177423,0.0008177576,0.0002508628,0.00130354,0.0008592666,0.1365626,0.03660092,0.006366613,0.005276224,0.7887459],"study_design_scores_gemma":[0.0001819519,0.0006295071,0.007209118,0.0003404552,0.0001896412,0.001131815,0.0002219808,0.893377,0.06869554,0.01838878,0.009519825,0.0001143745],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09726414,0.0008225106,0.869621,0.0005174869,0.00005613618,0.0004568061,0.0003810375,0.0279218,0.002959018],"genre_scores_gemma":[0.5099316,0.0004264998,0.4847881,0.0003355819,0.00004988772,0.0003514761,0.001665204,0.00150639,0.0009452137],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.007126146,"threshold_uncertainty_score":0.03768712,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2142266413","doi":"10.5555/998675.999415","title":"Using simulation to empirically investigate test coverage criteria based on statechart","year":2004,"lang":"en","type":"article","venue":"International Conference on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":81,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Computer science; Code coverage; Class (philosophy); Test (biology); Test case; Reliability engineering; Fault detection and isolation; Test strategy; Data mining; Machine learning; Artificial intelligence; Programming language; Software; Engineering","authors":[{"name":"Lionel Briand","is_ca":true},{"name":"Yvan Labiche","is_ca":true},{"name":"Y. Wang","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08716524489143154,"gpt":0.3492094312045713,"spread":0.2620441863131397,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01624715,0.0009200234,0.0007054651,0.003202317,0.0004137338,0.001521469,0.00106193,0.001380793,0.001719989],"category_scores_gemma":[0.1230567,0.0004647213,0.0009635579,0.002778782,0.001019205,0.002487482,0.0009162495,0.001040766,0.0001414236],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002767218,"about_ca_system_score_gemma":0.00134342,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004847192,"about_ca_topic_score_gemma":0.005274597,"domain_scores_codex":[0.9864815,0.009789949,0.0007214919,0.000617566,0.001873933,0.0005155331],"domain_scores_gemma":[0.6261046,0.353396,0.007147089,0.0067996,0.006010451,0.0005422574],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008560826,0.0006528771,0.05660278,0.0002729327,0.0003844672,0.0001144685,0.0004304375,0.8989832,0.001987103,0.01774184,0.0004017212,0.02157218],"study_design_scores_gemma":[0.0001006258,0.001243089,0.01072382,0.00006765802,0.0001399374,0.00008806279,0.0002654979,0.9747321,0.003218948,0.008781188,0.0006017736,0.00003743161],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.920634,0.0003506327,0.072308,0.0002797819,0.0000155279,0.0003514563,0.0005394383,0.0001734687,0.005347793],"genre_scores_gemma":[0.9843237,0.00008664277,0.01499774,0.00002025706,0.000003455101,0.0002076251,0.0002296073,0.00001275745,0.0001183195],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01624715,"threshold_uncertainty_score":0.08592421,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2100209048","doi":"","title":"Static And Dynamic Reverse Engineering Techniques for Java Software Systems","year":2019,"lang":"en","type":"book","venue":"Trepo - Institutional Repository of Tampere University","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":81,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Tekes; University of Victoria","keywords":"Reverse engineering; Computer science; Java; Software engineering; Programming language","authors":[{"name":"Tarja Systä","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.008589263300448567,"gpt":0.1962618870493534,"spread":0.1876726237489048,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00204157,0.0009313688,0.0004842676,0.001694147,0.0007625769,0.002583529,0.001694218,0.0007657107,0.007954699],"category_scores_gemma":[0.006072253,0.001072757,0.002130115,0.001225119,0.001249784,0.004002583,0.002798929,0.002589964,0.001857741],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009186727,"about_ca_system_score_gemma":0.001477943,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002324106,"about_ca_topic_score_gemma":0.004144958,"domain_scores_codex":[0.9967781,0.0006736633,0.0002994742,0.0005618327,0.001458774,0.0002281475],"domain_scores_gemma":[0.9970912,0.001289102,0.0002393307,0.0008302775,0.0005049584,0.00004506791],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009896105,0.0001289507,0.0009877925,0.0008137547,0.00007992761,0.0004917815,0.001407359,0.04023721,0.03523664,0.416142,0.004130787,0.5002449],"study_design_scores_gemma":[0.0001004939,0.0001946661,0.0009611298,0.0005075107,0.0002489138,0.001309174,0.0009495071,0.402169,0.099424,0.3021057,0.191918,0.0001119885],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.002151718,0.0001732548,0.9923434,0.00009354815,0.00002806875,0.0000467827,0.00003620974,0.0009100432,0.004217036],"genre_scores_gemma":[0.09187728,0.0007514514,0.8964073,0.0001503039,0.00004198252,0.0001703972,0.0003363825,0.0008737583,0.00939115],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.007954699,"threshold_uncertainty_score":0.02661109,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2068493323","doi":"10.1109/tse.2011.56","title":"Size-Constrained Regression Test Case Selection Using Multicriteria Optimization","year":2011,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":79,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Test suite; Test case; Consistency (knowledge bases); Integer programming; Mathematical optimization; Task (project management); Constraint (computer-aided design); Selection (genetic algorithm); Regression testing; Greedy algorithm; Linear programming; Relaxation (psychology); Software; Algorithm; Machine learning; Regression analysis; Artificial intelligence; Software system; Mathematics","authors":[{"name":"Siavash Mirarab","is_ca":false},{"name":"Soroush Akhlaghi","is_ca":false},{"name":"Ladan Tahvildari","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03160363037280706,"gpt":0.2469561588901617,"spread":0.2153525285173546,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003340231,0.001793222,0.002122707,0.002578,0.0004557037,0.0009586763,0.001941458,0.001037073,0.002263932],"category_scores_gemma":[0.01214569,0.0008152775,0.001388808,0.001491523,0.0008785232,0.001026046,0.00104371,0.001120161,0.0003972832],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001279627,"about_ca_system_score_gemma":0.001842395,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0040839,"about_ca_topic_score_gemma":0.004639899,"domain_scores_codex":[0.9963698,0.001880121,0.000155738,0.0004602536,0.0008076758,0.000326385],"domain_scores_gemma":[0.9894552,0.008016248,0.0009346336,0.0005006625,0.0009296461,0.0001635896],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001828485,0.0001576472,0.001433539,0.00008623061,0.00009261302,0.0001082331,0.00004997827,0.885776,0.005439363,0.002811615,0.0008575861,0.1030044],"study_design_scores_gemma":[0.00003017518,0.00006433009,0.0002140152,0.000005955796,0.00001546945,0.00002570279,0.000008331274,0.9969354,0.001384863,0.001139893,0.0001699522,0.000005929534],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04221788,0.0002500273,0.9535501,0.0001895655,0.00001864077,0.0002792964,0.00005645239,0.001129935,0.002307964],"genre_scores_gemma":[0.5088668,0.0001315805,0.4878107,0.0001968448,0.00003514525,0.000607582,0.0003522329,0.0003059103,0.001693203],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0040839,"threshold_uncertainty_score":0.01766503,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}