{"id":"W7105804309","doi":"10.2196/73725","title":"Iterative Large Language Model–Guided Sampling and Expert-Annotated Benchmark Corpus for Harmful Suicide Content Detection: Development and Validation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Mental Health via Writing","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Benchmark (surveying); Experience sampling method; Sampling (signal processing); Poison control; Moderation; Focus (optics); Human factors and ergonomics; Content (measure theory)","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01243356,0.001780487,0.001110978,0.003941612,0.002040439,0.001893221,0.003742244,0.002301008,0.005297601],"category_scores_gemma":[0.03851629,0.0006146389,0.001382656,0.002338824,0.001929523,0.002353972,0.003482095,0.002342544,0.005064125],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002025297,"about_ca_system_score_gemma":0.003961636,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01829646,"about_ca_topic_score_gemma":0.03109473,"domain_scores_codex":[0.9896747,0.005537285,0.0008245585,0.001983602,0.001606237,0.0003735233],"domain_scores_gemma":[0.9637042,0.01922366,0.001016405,0.004756222,0.01050834,0.0007912017],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001630675,0.003925148,0.05418869,0.005100275,0.0007634354,0.003506348,0.005853853,0.05086922,0.03937193,0.00647611,0.1873145,0.6409999],"study_design_scores_gemma":[0.001326329,0.00201931,0.07962482,0.001049308,0.0007305777,0.002905701,0.00657236,0.7057626,0.0454528,0.00848099,0.1456134,0.0004617761],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6357927,0.004028109,0.2586809,0.001641602,0.001382134,0.007470116,0.04960863,0.02140402,0.01999184],"genre_scores_gemma":[0.4560718,0.0006694929,0.3454336,0.00103061,0.0002573876,0.007137077,0.1800178,0.001433997,0.007948336],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01829646,"threshold_uncertainty_score":0.06575572,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1097945635355415,"score_gpt":0.4424164120151892,"score_spread":0.3326218484796477,"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."}}