{"id":"W1968799289","doi":"10.1145/2808194.2809493","title":"Pooling for User-Oriented Evaluation Measures","year":2015,"lang":"en","type":"article","venue":"","topic":"Data Management and Algorithms","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; University of Waterloo","keywords":"Pooling; Automatic summarization; Computer science; Information retrieval; Relevance (law); Focus (optics); Rank (graph theory); Order (exchange); World Wide Web; Data mining; Artificial intelligence","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001105527,0.0000482207,0.00004730545,0.0000526017,0.00004126926,0.000131359,0.0003512884,0.00001265046,0.000005740677],"category_scores_gemma":[0.000136486,0.00003990634,0.00001994522,0.0001550405,0.000005387473,0.0007564099,0.0001225458,0.00001658256,0.00003968177],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002479519,"about_ca_system_score_gemma":0.00003832116,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001609582,"about_ca_topic_score_gemma":0.000008190511,"domain_scores_codex":[0.9992406,0.00002261158,0.00008592287,0.0001769927,0.0003538498,0.000119983],"domain_scores_gemma":[0.9994058,0.00001834233,0.00002644262,0.0002647749,0.0002328731,0.00005176291],"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":[0.000006594822,0.00004806921,0.0002142947,0.000005550389,0.00002280855,7.406424e-7,0.0002753434,0.0004133211,0.00005238831,0.4516799,0.07796503,0.4693159],"study_design_scores_gemma":[0.0006607364,0.00004670299,0.0001674095,0.000003284199,0.00001026332,4.209045e-7,0.00006194198,0.822941,0.0005439413,0.01007155,0.1654035,0.00008925634],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0006281164,0.00003359859,0.9872723,0.0006456266,0.0005477719,0.0002667959,0.000001201744,0.0001242032,0.01048044],"genre_scores_gemma":[0.2367522,0.000002972045,0.7570118,0.0007646512,0.0003220549,0.000114932,0.00006561101,0.00001042992,0.004955352],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8225277,"threshold_uncertainty_score":0.1627333,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1201717971347841,"score_gpt":0.3264951012284474,"score_spread":0.2063233040936634,"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."}}