{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001887693,0.0002812549,0.000209416,0.0002650501,0.0005886757,0.0008230111,0.001415263,0.0002003135,0.0001622802],"category_scores_gemma":[0.00140959,0.0001938157,0.0001870856,0.0005363756,0.00004458378,0.004196519,0.0003362064,0.000368089,0.001213138],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001969559,"about_ca_system_score_gemma":0.0001889689,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001088983,"about_ca_topic_score_gemma":0.000007069863,"domain_scores_codex":[0.9965816,0.00008944537,0.0008154283,0.0003409873,0.001522876,0.0006496632],"domain_scores_gemma":[0.9962543,0.0003817397,0.0003536083,0.00161452,0.001236244,0.0001595755],"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":[0.0004651328,0.0006538851,0.001376987,0.00007603786,0.0001061191,0.00001365117,0.00374848,0.07092222,0.0008113203,0.05730484,0.3505681,0.5139532],"study_design_scores_gemma":[0.001082989,0.0004416369,0.003055407,0.00007029072,0.0000410658,0.00002923311,0.0004463021,0.9687613,0.002075276,0.005553945,0.01792267,0.000519836],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7289633,0.00003635065,0.2229654,0.0360594,0.001842129,0.007274549,0.00003015967,0.0004117062,0.002417053],"genre_scores_gemma":[0.9886406,0.00002769713,0.00523011,0.003467985,0.0001294911,0.001150347,0.0001963187,0.00002337937,0.001134085],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8978391,"threshold_uncertainty_score":0.9995645,"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."}}