{"id":"W4389523922","doi":"10.18653/v1/2023.findings-emnlp.928","title":"Learning to love diligent trolls: Accounting for rater effects in the dialogue safety task","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Chatbot; Utterance; Task (project management); Consistency (knowledge bases); Offensive; Artificial intelligence; Human–computer interaction; Natural language processing; Machine learning; Information retrieval; Operations research; Engineering","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02766536,0.001825117,0.001209584,0.001410786,0.001240931,0.003343514,0.001920667,0.002124177,0.002862794],"category_scores_gemma":[0.1087458,0.0009468411,0.0008488925,0.0009515614,0.001027951,0.004088002,0.003024969,0.004133919,0.002217761],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001134946,"about_ca_system_score_gemma":0.001302449,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008785086,"about_ca_topic_score_gemma":0.01045277,"domain_scores_codex":[0.9860381,0.008389219,0.000660125,0.003253255,0.001068044,0.0005912956],"domain_scores_gemma":[0.9066138,0.06899269,0.00633889,0.009923501,0.006155253,0.001975824],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00248042,0.001028484,0.240843,0.0006996159,0.0009505461,0.0007602585,0.007683308,0.1377322,0.02030027,0.007250353,0.01523421,0.5650374],"study_design_scores_gemma":[0.00005673113,0.0003403579,0.04085259,0.00008261528,0.0001444182,0.0002503002,0.0005946259,0.9390028,0.004514881,0.01057489,0.003491617,0.00009410179],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4520155,0.002408796,0.529739,0.001803915,0.0006372621,0.000513139,0.0009606935,0.003834484,0.008087158],"genre_scores_gemma":[0.9364418,0.0002082442,0.056252,0.0002833089,0.0003000745,0.0002794586,0.001023844,0.000390272,0.004821009],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9723346,"threshold_uncertainty_score":0.1463102,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01836665374351341,"score_gpt":0.2613305293758066,"score_spread":0.2429638756322932,"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."}}