{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009379397,0.0001730709,0.0001144129,0.0002902345,0.000742303,0.0005067751,0.001133301,0.00005168262,0.00005601884],"category_scores_gemma":[0.0002523222,0.000149347,0.0000567878,0.0007780606,0.00001402932,0.0008732146,0.00007930736,0.0003477125,0.0001203012],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002074515,"about_ca_system_score_gemma":0.00006591585,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009639153,"about_ca_topic_score_gemma":0.0001083803,"domain_scores_codex":[0.9983237,0.0001245326,0.0002975797,0.0006825408,0.0003118733,0.0002598028],"domain_scores_gemma":[0.9980989,0.0003784693,0.00007005772,0.001296764,0.00005930321,0.0000964994],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004950673,0.00002621088,0.000001820383,0.00002908835,0.00003644472,4.548862e-7,0.001196393,0.2513371,0.05837989,0.0004765557,0.0004053351,0.6880612],"study_design_scores_gemma":[0.0001643909,0.0002550711,0.0000310963,0.0001007819,0.0000267306,0.000007119275,0.0001945479,0.9225933,0.05114526,0.0001944926,0.02509885,0.000188346],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01174878,0.00001828271,0.9732627,0.01296699,0.0007946905,0.0006104503,0.00001374783,0.0004393369,0.0001449806],"genre_scores_gemma":[0.9316815,0.00002282069,0.06622511,0.0004466479,0.00009545508,0.0002472094,0.00006422222,0.00002046115,0.001196607],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9199327,"threshold_uncertainty_score":0.6090196,"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."}}