{"id":"W4394844693","doi":"10.1371/journal.pdig.0000489","title":"Biomedical text readability after hypernym substitution with fine-tuned large language models","year":2024,"lang":"en","type":"article","venue":"PLOS Digital Health","topic":"Text Readability and Simplification","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; University of Toronto","funders":"","keywords":"Readability; Computer science; Substitution (logic); Unified Medical Language System; Natural language processing; Text simplification; Automatic summarization; Artificial intelligence; Sentence; Comprehension; Biomedical text mining; Language model; Text mining","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":[0.004206232,0.00113503,0.000695698,0.001285596,0.0003435329,0.001619446,0.0007484237,0.0006648403,0.002194659],"category_scores_gemma":[0.03536383,0.000290701,0.0009511445,0.0007699024,0.0004714869,0.002058921,0.001245134,0.001687928,0.001051838],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007982586,"about_ca_system_score_gemma":0.0007634651,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004278324,"about_ca_topic_score_gemma":0.005874621,"domain_scores_codex":[0.9966342,0.001792234,0.0003672289,0.0006973639,0.0003987114,0.000110237],"domain_scores_gemma":[0.9679673,0.02508925,0.001470551,0.002705697,0.002436568,0.0003305743],"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":[0.004767673,0.001200994,0.03141816,0.001737639,0.0008031549,0.0009091608,0.005288464,0.1337732,0.1787473,0.001973226,0.00942062,0.6299604],"study_design_scores_gemma":[0.00023193,0.001586819,0.04356626,0.0001302636,0.0003746197,0.0004380537,0.001943718,0.8241662,0.1117965,0.006094536,0.009430156,0.0002409328],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8385175,0.0007553215,0.1429494,0.0005809301,0.0002115891,0.0003548796,0.003009767,0.01142361,0.002196959],"genre_scores_gemma":[0.8735939,0.0001709634,0.1177895,0.0001501714,0.00004088996,0.0003067737,0.006097603,0.0006276165,0.001222558],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004278324,"threshold_uncertainty_score":0.02224493,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01983503389309309,"score_gpt":0.2663777210944307,"score_spread":0.2465426872013376,"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."}}