{"id":"W3217062054","doi":"10.3390/math9222988","title":"Can Fake News Detection Models Maintain the Performance through Time? A Longitudinal Evaluation of Twitter Publications","year":2021,"lang":"en","type":"article","venue":"Mathematics","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Robustness (evolution); Computer science; Context (archaeology); Word embedding; Machine learning; Feature engineering; Focus (optics); Artificial intelligence; Longitudinal data; Term (time); Feature (linguistics); Embedding; Data science; Deep learning; Data mining","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":[0.001456159,0.00006590066,0.00009851252,0.000036765,0.0003542713,0.0001078949,0.0001449153,0.000052332,0.0006963217],"category_scores_gemma":[0.0005238323,0.00004956168,0.00004550927,0.0004292537,0.00008965611,0.0004764459,0.00002686727,0.00007587061,0.00004352513],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001199876,"about_ca_system_score_gemma":0.0003594147,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000108599,"about_ca_topic_score_gemma":0.000800047,"domain_scores_codex":[0.9986422,0.0001358229,0.0002727846,0.00007803788,0.0007130778,0.0001580874],"domain_scores_gemma":[0.9988579,0.00008909048,0.0001826706,0.0002647402,0.0005646814,0.00004095183],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000007507737,0.0003060536,0.0003437113,0.0002156276,0.0001112269,3.385731e-7,0.6239437,0.009655362,0.0006422981,0.3100713,0.005741662,0.04896115],"study_design_scores_gemma":[0.0006372881,0.00004598498,0.002600216,0.0001207114,0.000198638,0.00001747299,0.09220517,0.7253373,0.005735829,0.163492,0.009314114,0.0002952175],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8062544,0.00003735743,0.03168831,0.008328482,0.0001019845,0.0005018137,0.000007066353,0.00005628913,0.1530243],"genre_scores_gemma":[0.9954278,0.00004779781,0.00212562,0.000202877,0.00006749226,0.00001369321,0.000008422989,0.000005972072,0.002100267],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.715682,"threshold_uncertainty_score":0.7624235,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1535578308925872,"score_gpt":0.3610018866829584,"score_spread":0.2074440557903712,"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."}}