{"id":"W2029009258","doi":"10.1145/2348283.2348300","title":"Time-based calibration of effectiveness measures","year":2012,"lang":"en","type":"article","venue":"","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":200,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"University of Waterloo","keywords":"Computer science; Measure (data warehouse); Process (computing); Quality (philosophy); Information retrieval; Calibration; Adaptation (eye); Data mining; Machine learning","routes":{"ca_aff":true,"ca_fund":true,"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":[0.03200497,0.001284413,0.0008742473,0.004797782,0.0006308739,0.002321976,0.002352586,0.001884615,0.002867628],"category_scores_gemma":[0.3036897,0.000562711,0.0009430076,0.004493567,0.001479991,0.004468923,0.001878044,0.001663689,0.001291005],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002473348,"about_ca_system_score_gemma":0.0009465561,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002536812,"about_ca_topic_score_gemma":0.001923103,"domain_scores_codex":[0.9519276,0.025659,0.003170342,0.004704596,0.01364213,0.0008961958],"domain_scores_gemma":[0.687858,0.2212814,0.01767446,0.04321877,0.02863112,0.001336238],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003300424,0.001850441,0.1190937,0.002258078,0.001137494,0.0002782678,0.003554696,0.2064402,0.0266499,0.05137167,0.01640589,0.5676592],"study_design_scores_gemma":[0.0004452246,0.005811258,0.2162781,0.0008344466,0.0006882705,0.001690671,0.001743121,0.5894116,0.07379001,0.06165162,0.04695525,0.000700498],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3315251,0.003774985,0.6272882,0.0006738813,0.0003910474,0.001648501,0.002825499,0.002776228,0.02909657],"genre_scores_gemma":[0.846961,0.0004377394,0.1458361,0.0002970345,0.0001169222,0.001587576,0.002380563,0.0006388359,0.001744143],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.03200497,"threshold_uncertainty_score":0.1692604,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02563785066914584,"score_gpt":0.2660673763852607,"score_spread":0.2404295257161148,"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."}}