{"id":"W4403850457","doi":"10.1145/3700791","title":"AugmenToxic: Leveraging Reinforcement Learning to Optimize LLM Instruction Fine-Tuning for Data Augmentation to Enhance Toxicity Detection","year":2024,"lang":"en","type":"article","venue":"ACM Transactions on the Web","topic":"Machine Learning and Data Classification","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Concordia University","funders":"Canada Research Chairs","keywords":"Computer science; Reinforcement learning; Human–computer interaction; Machine learning","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003025711,0.001794623,0.001099881,0.0007973664,0.0005220574,0.001177343,0.002016694,0.001319462,0.002208649],"category_scores_gemma":[0.0151107,0.0004924214,0.0009149454,0.0005455202,0.001060463,0.002368849,0.002235638,0.003129183,0.001662307],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001109552,"about_ca_system_score_gemma":0.001784921,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003201929,"about_ca_topic_score_gemma":0.005734005,"domain_scores_codex":[0.9979133,0.0008459992,0.0001036295,0.000685922,0.0002884037,0.0001627763],"domain_scores_gemma":[0.9959485,0.002298611,0.0002563879,0.0007288338,0.0005622074,0.0002056064],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008804146,0.001268212,0.01547034,0.0004790161,0.0001868847,0.0003318766,0.0005572279,0.4654787,0.03103706,0.004831448,0.01558451,0.4638943],"study_design_scores_gemma":[0.00004772318,0.0001851649,0.0006727609,0.00001866001,0.00001881239,0.00003712824,0.00004992135,0.984138,0.008262602,0.004304545,0.002239778,0.00002494577],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1791989,0.001193429,0.7913982,0.001247844,0.000332618,0.00052718,0.001683738,0.02045087,0.003967281],"genre_scores_gemma":[0.7087837,0.0001927923,0.2807522,0.001084462,0.0001119318,0.0008119547,0.004199518,0.0007654119,0.00329792],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.003201929,"threshold_uncertainty_score":0.0160017,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04766722686026172,"score_gpt":0.3230694024013829,"score_spread":0.2754021755411212,"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."}}