{"id":"W1501070457","doi":"10.1023/a:1025603704680","title":"Current Status of the Evaluation of Information Retrieval","year":2003,"lang":"en","type":"review","venue":"Journal of Medical Systems","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":40,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Computer science; Relevance (law); Information retrieval; Data science; Management science; Engineering","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":[0.02250004,0.001311343,0.004951612,0.009440516,0.0008316221,0.006555294,0.003226205,0.003670231,0.005006392],"category_scores_gemma":[0.05872265,0.0006746096,0.0009667709,0.01216832,0.002813997,0.007208201,0.001424019,0.001996787,0.002201839],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003613037,"about_ca_system_score_gemma":0.007605376,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00717263,"about_ca_topic_score_gemma":0.009064627,"domain_scores_codex":[0.989408,0.003935291,0.001394575,0.0008042275,0.004184828,0.0002730222],"domain_scores_gemma":[0.887382,0.08966409,0.003694289,0.001410228,0.01679237,0.001056972],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001252769,0.00009412078,0.001010122,0.01137428,0.0001050036,0.00002561804,0.00009065139,0.0002603923,0.0001584726,0.003836053,0.01577571,0.9671443],"study_design_scores_gemma":[0.0001530954,0.0006625564,0.01250611,0.04742452,0.001628397,0.001295374,0.001111535,0.003539889,0.001938861,0.03653751,0.8930011,0.0002010364],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000223953,0.9961919,0.00087027,0.001376094,0.0001630965,0.00001016741,0.00002063229,0.00001609998,0.001127806],"genre_scores_gemma":[0.004909965,0.9902827,0.002526086,0.0007395472,0.0009123828,0.00003013303,0.00006988816,0.00001354088,0.0005158882],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9775,"threshold_uncertainty_score":0.118993,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1294945269002648,"score_gpt":0.4170239174674541,"score_spread":0.2875293905671893,"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."}}