{"id":"W2158222053","doi":"10.1145/2568388.2568403","title":"Report on the SIGIR 2013 workshop on modeling user behavior for information retrieval evaluation (MUBE 2013)","year":2013,"lang":"en","type":"article","venue":"ACM SIGIR Forum","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of Waterloo","funders":"","keywords":"Breakout; Computer science; Set (abstract data type); Information retrieval; World Wide Web","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04906387,0.002494968,0.002796731,0.001944748,0.002126455,0.008848806,0.003356149,0.004053543,0.03692502],"category_scores_gemma":[0.06815179,0.001458453,0.00300515,0.002031242,0.000989042,0.01097488,0.005231034,0.006423253,0.0208503],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003905623,"about_ca_system_score_gemma":0.004800399,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0216201,"about_ca_topic_score_gemma":0.02410958,"domain_scores_codex":[0.9755039,0.0130075,0.0008901902,0.00226575,0.007313593,0.001019067],"domain_scores_gemma":[0.9477548,0.02432152,0.001159798,0.006694922,0.01642682,0.003642199],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0007179984,0.001080545,0.004686919,0.0007062954,0.0002350514,0.0001261068,0.001035944,0.007216298,0.00295487,0.0122128,0.7134224,0.2556048],"study_design_scores_gemma":[0.0004039972,0.001054275,0.01030005,0.001057439,0.0004928717,0.0003393411,0.001909684,0.08542905,0.01212195,0.05896081,0.827483,0.0004475754],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.03049997,0.03017804,0.7450169,0.06792738,0.01977717,0.004616065,0.02317345,0.01062336,0.0681877],"genre_scores_gemma":[0.1622614,0.02509285,0.4679898,0.02019272,0.008379394,0.007697383,0.08999802,0.007102834,0.2112856],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9509361,"threshold_uncertainty_score":0.2594777,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05571689524022985,"score_gpt":0.3110163909179287,"score_spread":0.2552994956776988,"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."}}