{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01259633,0.0001569821,0.001012041,0.0003314651,0.00004308505,0.00007585051,0.001535724,0.0002666492,0.00004023259],"category_scores_gemma":[0.002524662,0.00007969355,0.00050352,0.0009204773,0.00007394355,0.0007313043,0.0001521954,0.0007572866,0.00001060177],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003378148,"about_ca_system_score_gemma":0.005543263,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001150903,"about_ca_topic_score_gemma":1.677787e-7,"domain_scores_codex":[0.9868911,0.001195568,0.002915497,0.00006755539,0.008698542,0.0002317717],"domain_scores_gemma":[0.9925904,0.0002175119,0.004501062,0.000362865,0.002076544,0.000251588],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000002902691,0.00003697419,0.000006796908,0.006885944,0.00003884011,4.776189e-7,0.0002143821,0.00001298206,6.683418e-8,0.002162114,0.0008634372,0.9897751],"study_design_scores_gemma":[0.0004590047,0.00009741155,0.00003581066,0.01584484,0.0002559707,0.0001654458,0.00003659549,0.002962241,0.000002771628,0.00002515202,0.980031,0.00008374576],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00005645432,0.9821607,0.01204525,0.00003702417,0.004837697,0.0006535292,0.00001487461,0.000003726513,0.0001907518],"genre_scores_gemma":[0.0003018201,0.9994839,0.00006972748,0.000007183465,0.0001165541,0.000004012658,0.000004580345,0.000003909941,0.000008356212],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9896913,"threshold_uncertainty_score":0.9833511,"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."}}