{"id":"W4200056214","doi":"10.3389/fdgth.2021.801204","title":"Developing a Quality Benchmark for Determining the Credibility of Web Health Information- a Protocol of a Gold Standard Approach","year":2021,"lang":"en","type":"article","venue":"Frontiers in Digital Health","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Computer science; Credibility; Quality (philosophy); Misinformation; Knowledge management; Protocol (science); Data science; Medicine; Computer security","routes":{"ca_aff":true,"ca_fund":true,"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":[0.003528646,0.00009099951,0.0003811845,0.00009908428,0.0002090299,0.0001028944,0.0002066402,0.00005745832,0.000004521079],"category_scores_gemma":[0.00226582,0.00007395054,0.00007530914,0.0005484547,0.0001795006,0.001254681,0.00004661034,0.0001031636,2.457758e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000559038,"about_ca_system_score_gemma":0.007811555,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002266162,"about_ca_topic_score_gemma":0.0002838632,"domain_scores_codex":[0.9977335,0.0001754318,0.001111452,0.0001123458,0.0005028318,0.0003643784],"domain_scores_gemma":[0.9983792,0.0001522599,0.0007473622,0.0002040774,0.0003791001,0.0001379862],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0004909925,0.0002570229,0.05912814,0.006107346,0.00003347031,7.510138e-8,0.2472749,0.00007272899,1.443095e-7,0.04031599,0.04670518,0.599614],"study_design_scores_gemma":[0.008750719,0.001096119,0.1181402,0.001218374,0.000006185514,0.000001829549,0.4424715,0.005395516,0.00006939512,0.01965262,0.40249,0.0007075467],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1447356,0.0002343289,0.4709877,0.01618764,0.001378306,0.2278587,0.003184932,0.0001800638,0.1352526],"genre_scores_gemma":[0.8945628,0.00004398712,0.09800097,0.002553354,0.00006946488,0.004382174,0.0002179318,0.00001343294,0.0001558381],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7498272,"threshold_uncertainty_score":0.9978132,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06915478684436611,"score_gpt":0.4106840072210206,"score_spread":0.3415292203766545,"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."}}