{"id":"W2587282850","doi":"10.29173/cais44","title":"What are the Important Variables in the Evaluation of Information Retrieval Systems?","year":2013,"lang":"en","type":"article","venue":"Proceedings of the Annual Conference of CAIS / Actes du congrès annuel de l ACSI","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Computer science; Information retrieval; Precision and recall; Recall; Order (exchange); Data science; Psychology; Cognitive psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.06390972,0.001913897,0.003987187,0.00952424,0.001722819,0.01487508,0.001933089,0.00382879,0.003944283],"category_scores_gemma":[0.3111291,0.000906586,0.001343148,0.0142614,0.006343368,0.02185418,0.001519741,0.004523261,0.001827083],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004183283,"about_ca_system_score_gemma":0.002048686,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003090531,"about_ca_topic_score_gemma":0.002495328,"domain_scores_codex":[0.9137542,0.05823326,0.006273208,0.00331834,0.01692024,0.001500815],"domain_scores_gemma":[0.4391973,0.5013075,0.02784162,0.004371539,0.02431225,0.002969836],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00314241,0.0009455159,0.2209563,0.008645403,0.002527593,0.0002847465,0.001309539,0.01142771,0.002206801,0.05489811,0.02453711,0.6691187],"study_design_scores_gemma":[0.0009594508,0.004027109,0.5365044,0.00781309,0.003201365,0.0009980864,0.004706439,0.06749344,0.008647802,0.3057019,0.05887544,0.001071572],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.2477835,0.4302788,0.1712409,0.06776831,0.004580438,0.001514666,0.005358242,0.001206171,0.07026891],"genre_scores_gemma":[0.915979,0.02610843,0.04359069,0.003030014,0.005062678,0.001316817,0.001348748,0.0004544279,0.003109223],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9360903,"threshold_uncertainty_score":0.3379909,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03583881779892486,"score_gpt":0.2687064287646606,"score_spread":0.2328676109657357,"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."}}