{"id":"W7116920285","doi":"10.1080/10447318.2025.2597504","title":"Evaluation Metrics for Misinformation Warning Interventions: Challenges and Prospects","year":2025,"lang":"en","type":"article","venue":"International Journal of Human-Computer Interaction","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University","funders":"","keywords":"Misinformation; Work (physics); Warning system; The Internet; Key (lock)","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":[0.0033173,0.00009027753,0.0001518145,0.001004409,0.0002250565,0.0003441073,0.0002413684,0.00006936053,0.00007044538],"category_scores_gemma":[0.001111516,0.0000871236,0.0001624228,0.0001272004,0.00004206679,0.002206874,0.00003755583,0.0001527819,0.000004389571],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000489463,"about_ca_system_score_gemma":0.0001559225,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002249775,"about_ca_topic_score_gemma":0.00007992255,"domain_scores_codex":[0.9981465,0.0001406739,0.000707145,0.00009095902,0.0007957098,0.0001190476],"domain_scores_gemma":[0.9962202,0.0002277807,0.0007594555,0.00006116635,0.002674263,0.00005714198],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007292364,0.0001032536,0.00004468967,0.00006315878,0.0002581743,8.391305e-7,0.02503297,0.0002877543,0.00007099134,0.04480442,0.003232427,0.9260284],"study_design_scores_gemma":[0.02048378,0.002927297,0.06184292,0.01309129,0.001393004,0.0002890103,0.1076182,0.1898685,0.004468009,0.1430387,0.4534842,0.00149499],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4092536,0.009208906,0.4450748,0.01893248,0.02705906,0.002295319,0.00001255681,0.0001603406,0.08800299],"genre_scores_gemma":[0.9952312,0.0009576336,0.002516575,0.0002416025,0.0007544667,0.000006259113,0.00001050249,0.000005031419,0.0002767257],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9245334,"threshold_uncertainty_score":0.3552797,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1485662190976892,"score_gpt":0.4706139314356272,"score_spread":0.322047712337938,"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."}}