{"id":"W1973458428","doi":"10.1023/b:joms.0000021524.90614.a9","title":"A New Look at Information Retrieval Evaluation: Proposal for Solutions","year":2004,"lang":"en","type":"article","venue":"Journal of Medical Systems","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Victoria","funders":"Medical Research Council","keywords":"Relevance (law); Computer science; Terminology; Information retrieval; Identification (biology); Precision and recall; Information system; Data science; Management science; Data mining","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":[0.03477013,0.002299678,0.005201039,0.009280303,0.004234458,0.02126107,0.006678278,0.01366986,0.009735106],"category_scores_gemma":[0.09665501,0.001743892,0.002941553,0.007820268,0.01431916,0.03950539,0.005239654,0.01245515,0.002428095],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007727501,"about_ca_system_score_gemma":0.005980684,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006688059,"about_ca_topic_score_gemma":0.005542711,"domain_scores_codex":[0.9694926,0.01506444,0.002503361,0.0036738,0.008138761,0.001127081],"domain_scores_gemma":[0.9282366,0.04717507,0.002304901,0.00501176,0.01519006,0.00208157],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002654077,0.0003419881,0.002986803,0.0008017487,0.0002695683,0.0001406527,0.0007666188,0.004245799,0.0006638387,0.7561049,0.04776493,0.1856477],"study_design_scores_gemma":[0.00008444554,0.00008969519,0.000692663,0.0002812627,0.00009781188,0.0001933282,0.00053986,0.03351633,0.0004138321,0.9490865,0.01490044,0.0001036794],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005443561,0.02434533,0.7495852,0.1992877,0.001848526,0.0003896955,0.0003186106,0.0008937605,0.01788765],"genre_scores_gemma":[0.2674438,0.01198494,0.6658793,0.02296741,0.01105843,0.001347229,0.0006486512,0.0007751875,0.01789501],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9652299,"threshold_uncertainty_score":0.1838842,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04250847837540634,"score_gpt":0.3186410705850684,"score_spread":0.2761325922096621,"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."}}