{"id":"W3152534132","doi":"10.1145/3404835.3462947","title":"Evaluation Measures Based on Preference Graphs","year":2021,"lang":"en","type":"article","venue":"","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Ranking (information retrieval); Preference; Relevance (law); Measure (data warehouse); Computer science; Learning to rank; Similarity (geometry); Flexibility (engineering); Rank (graph theory); Set (abstract data type); Information retrieval; Similarity measure; Mathematics; Artificial intelligence; Statistics; Data mining; Combinatorics","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":[0.02053851,0.002855183,0.002362868,0.01026069,0.001060432,0.00619203,0.002369753,0.002110649,0.00460217],"category_scores_gemma":[0.09566023,0.0007286369,0.001980785,0.008048781,0.002064556,0.01097919,0.002475553,0.002951581,0.001131561],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004262201,"about_ca_system_score_gemma":0.001667083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001729442,"about_ca_topic_score_gemma":0.001559766,"domain_scores_codex":[0.9559177,0.0222584,0.003948952,0.003762425,0.01307109,0.001041383],"domain_scores_gemma":[0.8891134,0.08302213,0.006927215,0.007429245,0.01208823,0.001419794],"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":[0.001072611,0.00045777,0.007736776,0.001646946,0.0009332975,0.0002494251,0.0006655312,0.2192471,0.003607819,0.3020215,0.01246756,0.4498937],"study_design_scores_gemma":[0.0001673721,0.0008218428,0.003784246,0.0003084154,0.0002608743,0.0004144189,0.0003064317,0.5592418,0.003455731,0.420516,0.0105024,0.0002205403],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02380198,0.002510891,0.9592364,0.0007077272,0.0002001166,0.0007418885,0.001022613,0.0008937996,0.01088457],"genre_scores_gemma":[0.5722615,0.00175519,0.4177962,0.0005290007,0.0004755046,0.001692584,0.00182085,0.0003792018,0.00329009],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02053851,"threshold_uncertainty_score":0.1086193,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1327439096851362,"score_gpt":0.3104968045635845,"score_spread":0.1777528948784483,"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."}}