{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":["scholarly_communication"],"category_scores_codex":[0.004792891,0.0001671077,0.0002868226,0.0001815876,0.0001341976,0.005195485,0.002819295,0.0001114587,0.00001319369],"category_scores_gemma":[0.01012965,0.00008665645,0.0001020172,0.0009067166,0.0002227631,0.03620886,0.0004153095,0.0002640444,0.000004829863],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007017076,"about_ca_system_score_gemma":0.0003706233,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003067936,"about_ca_topic_score_gemma":0.000005080412,"domain_scores_codex":[0.9969868,0.0001021444,0.0009150615,0.0001380146,0.001572344,0.0002856119],"domain_scores_gemma":[0.9425331,0.0002001088,0.001612332,0.0003530695,0.05524796,0.00005343348],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0002562922,0.0004340647,0.2177432,0.001623811,0.000143991,7.805385e-7,0.4158007,0.0004332223,0.009925031,0.3113281,0.006404085,0.03590677],"study_design_scores_gemma":[0.001547653,0.0004489764,0.7287076,0.001149383,0.0001456045,0.00006362795,0.08439603,0.1428056,0.0195547,0.01395363,0.006758016,0.0004692638],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9933766,0.0001488994,0.0002015325,0.00332759,0.0002666176,0.00120053,0.00002776519,0.00001755519,0.001432872],"genre_scores_gemma":[0.9994392,0.0001004724,0.0001352819,0.0001943357,0.0000257938,0.00006123386,0.000003629993,0.000004044468,0.00003597705],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5109644,"threshold_uncertainty_score":0.9982085,"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."}}