{"id":"W4386805215","doi":"10.1016/j.jbi.2023.104486","title":"A self-supervised language model selection strategy for biomedical question answering","year":2023,"lang":"en","type":"article","venue":"Journal of Biomedical Informatics","topic":"Topic Modeling","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":false,"ca_institutions":"Toronto Metropolitan University; University of Waterloo","funders":"","keywords":"Computer science; Leverage (statistics); Artificial intelligence; Classifier (UML); Machine learning; Language model; Question answering; Retraining; Domain (mathematical analysis); Task (project management); Natural language processing","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":[0.001417232,0.0001138322,0.0002245653,0.0004794608,0.00008636373,0.0001223626,0.0005372379,0.0001516618,0.000003794533],"category_scores_gemma":[0.0001840596,0.00009071628,0.0001115061,0.0007175882,0.00003737069,0.0008660128,0.00009829514,0.000263528,0.00001247804],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009669299,"about_ca_system_score_gemma":0.0003262295,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002314626,"about_ca_topic_score_gemma":7.41809e-7,"domain_scores_codex":[0.9979686,0.00002751916,0.0009286385,0.00007853229,0.0006910308,0.0003056775],"domain_scores_gemma":[0.998966,0.0001107157,0.0003215549,0.0001560848,0.0001841789,0.000261502],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00009489865,0.0006978292,0.00009093797,0.002093787,0.0003846738,0.000102542,0.05036769,0.1051915,0.009450867,0.03430999,0.02171929,0.775496],"study_design_scores_gemma":[0.0006546374,0.0002879021,0.00001227007,0.00006875123,0.00001398067,0.0001081357,0.0004973382,0.9949064,0.0001782389,0.001457113,0.0017155,0.00009969163],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05704214,0.00002342796,0.9411749,0.0009710643,0.0004109234,0.000118485,0.000003849692,0.0001773321,0.00007790003],"genre_scores_gemma":[0.3278493,0.00005518823,0.6713935,0.0002529346,0.0003783855,0.000006509265,0.00001137377,0.00001022509,0.0000425703],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.889715,"threshold_uncertainty_score":0.3699302,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02976819324708699,"score_gpt":0.3077633349504817,"score_spread":0.2779951417033948,"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."}}