{"id":"W4400399317","doi":"10.1101/2024.07.05.24309412","title":"OQA : A question-answering dataset on orthodontic literature","year":2024,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal; Université de Montréal","funders":"","keywords":"Question answering; Information retrieval; Computer science","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":[],"consensus_categories":[],"category_scores_codex":[0.002049575,0.001039949,0.0006388458,0.005668529,0.0007488789,0.001066398,0.001789007,0.002425631,0.01143709],"category_scores_gemma":[0.01236118,0.0002828168,0.001224307,0.003721636,0.0005495431,0.001100766,0.00217716,0.001007872,0.007058293],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001217042,"about_ca_system_score_gemma":0.002216588,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009735346,"about_ca_topic_score_gemma":0.01786818,"domain_scores_codex":[0.9985877,0.0003867976,0.0002849023,0.0003521094,0.0002979106,0.00009054066],"domain_scores_gemma":[0.9942893,0.003458095,0.0004095966,0.0006107982,0.0008994822,0.0003326282],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001249516,0.0007443439,0.02029824,0.0109209,0.0002783212,0.001241581,0.001286887,0.00859943,0.01543351,0.003303279,0.7128329,0.223811],"study_design_scores_gemma":[0.0009844364,0.0008201078,0.08183191,0.001432387,0.0002879487,0.002526206,0.001995697,0.0475409,0.01475996,0.007817053,0.8398073,0.0001961384],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.07436412,0.007994188,0.01315818,0.002862426,0.000445606,0.001123781,0.8830913,0.009718047,0.007242373],"genre_scores_gemma":[0.06715475,0.001313371,0.03272334,0.0005755296,0.0001687258,0.001059038,0.8930324,0.0002176766,0.003755214],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.01143709,"threshold_uncertainty_score":0.03826088,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03006580840533227,"score_gpt":0.3026635983425187,"score_spread":0.2725977899371864,"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."}}