{"id":"W2407086192","doi":"10.2196/mental.4822","title":"Validating Machine Learning Algorithms for Twitter Data Against Established Measures of Suicidality","year":2016,"lang":"en","type":"article","venue":"JMIR Mental Health","topic":"Mental Health via Writing","field":"Psychology","cited_by":199,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Machine learning; Algorithm; Artificial intelligence; Computer science; Social media; Predictive value; Suicide prevention; Poison control; Medicine; Medical emergency; World Wide Web","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":[0.01270285,0.0008700884,0.000554634,0.00334108,0.000735276,0.001845466,0.0007662792,0.001228815,0.001493535],"category_scores_gemma":[0.06977354,0.0002093449,0.0006732858,0.001671096,0.0007189658,0.001893839,0.00116161,0.001079147,0.001357343],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009646523,"about_ca_system_score_gemma":0.0009743868,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002815636,"about_ca_topic_score_gemma":0.002455012,"domain_scores_codex":[0.9897654,0.006222377,0.001266183,0.001027226,0.001461177,0.0002577151],"domain_scores_gemma":[0.9494238,0.03276967,0.005008381,0.004592085,0.007695599,0.000510585],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001417262,0.001597875,0.7658953,0.0004992168,0.0007708579,0.0002590058,0.0008180393,0.06070235,0.004007061,0.002408047,0.009649275,0.1519756],"study_design_scores_gemma":[0.0002064127,0.001486526,0.1885593,0.000346143,0.0001622995,0.0004538596,0.00143945,0.7849684,0.01136424,0.006208683,0.004719339,0.00008534859],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.957132,0.0004180586,0.03229557,0.001180153,0.0002461058,0.0006030102,0.00359716,0.0006075698,0.003920438],"genre_scores_gemma":[0.9675639,0.0001324487,0.02769171,0.0001819604,0.00008513984,0.0003539998,0.003474985,0.0000324707,0.0004832783],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01270285,"threshold_uncertainty_score":0.06717992,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1912130276026938,"score_gpt":0.4545913416383931,"score_spread":0.2633783140356993,"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."}}