{"id":"W4412877107","doi":"10.1145/3711896.3737437","title":"A Guide to Misinformation Detection Data and Evaluation","year":2025,"lang":"en","type":"article","venue":"","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; McMaster University; Mila - Quebec Artificial Intelligence Institute; Vector Institute; Université de Montréal","funders":"","keywords":"Misinformation; Computer science; Computer security","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":[0.05026407,0.001997716,0.001596677,0.0129671,0.002094779,0.008756751,0.004719805,0.003970271,0.04629523],"category_scores_gemma":[0.2176322,0.002554378,0.001745309,0.01190999,0.001926637,0.007981163,0.005866671,0.005324985,0.04846354],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004111052,"about_ca_system_score_gemma":0.00955258,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007979988,"about_ca_topic_score_gemma":0.01607374,"domain_scores_codex":[0.9553514,0.02148392,0.007768331,0.002388041,0.01221569,0.0007926506],"domain_scores_gemma":[0.7872819,0.1162567,0.007356211,0.04037644,0.0462245,0.00250421],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001841199,0.0002742652,0.002467549,0.002039367,0.00008143319,0.000111914,0.0005111985,0.003166174,0.001026419,0.03163663,0.8471486,0.1113524],"study_design_scores_gemma":[0.0002422952,0.0001817948,0.003245895,0.00280517,0.00003612238,0.0002637501,0.0005232191,0.009713742,0.003669,0.05640609,0.9227563,0.0001567576],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"review","genre_scores_codex":[0.005532554,0.004263713,0.2663326,0.01083632,0.0009313281,0.01300614,0.6153835,0.03090139,0.05281253],"genre_scores_gemma":[0.01315334,0.002306758,0.5913851,0.004763325,0.000283262,0.02997643,0.3363962,0.006095588,0.01564],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.05026407,"threshold_uncertainty_score":0.265825,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06724764744455006,"score_gpt":0.4494331278618642,"score_spread":0.3821854804173141,"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."}}