{"id":"W3034927159","doi":"10.1145/3397271.3401412","title":"Vis-Trec: A System for the In-depth Analysis of trec_eval Results","year":2020,"lang":"en","type":"article","venue":"","topic":"Scientific Computing and Data Management","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University","funders":"Agence Nationale de la Recherche","keywords":"Computer science; Python (programming language); Information retrieval; Extensibility; Open source; Source code; World Wide Web; Programming language; Software","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009539312,0.00309492,0.001594804,0.008312948,0.001296603,0.005896542,0.00276927,0.00153837,0.0553978],"category_scores_gemma":[0.02860985,0.001450163,0.002221535,0.004588684,0.0007193585,0.005130593,0.004053333,0.00297613,0.04435613],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002103577,"about_ca_system_score_gemma":0.00466495,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01107723,"about_ca_topic_score_gemma":0.01365375,"domain_scores_codex":[0.9938533,0.001419959,0.0006377542,0.001244404,0.002350635,0.0004938082],"domain_scores_gemma":[0.9828153,0.006322828,0.001248615,0.003933171,0.004784832,0.0008952086],"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":[0.0009927928,0.0002568306,0.004788361,0.001190376,0.0003737585,0.0001603362,0.0005221539,0.004281276,0.00608128,0.003594507,0.8708375,0.1069209],"study_design_scores_gemma":[0.0008699179,0.0004440403,0.01690778,0.0007452044,0.000256263,0.0005143458,0.0005679481,0.142494,0.04596447,0.02408735,0.7661929,0.0009556453],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"software","genre_gemma":"software","genre_scores_codex":[0.005562228,0.0005955721,0.07197292,0.0005296444,0.0004880753,0.0004943794,0.1136281,0.7930368,0.01369233],"genre_scores_gemma":[0.06101947,0.0006608198,0.2578337,0.001040746,0.0004694269,0.002108564,0.4119759,0.2509285,0.01396296],"genre_candidate":"software","genre_consensus":"software","teacher_disagreement_score":0.0553978,"threshold_uncertainty_score":0.1853241,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2569924804930374,"score_gpt":0.4157226528767722,"score_spread":0.1587301723837348,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}