{"id":"W4285211372","doi":"10.18653/v1/2022.acl-long.378","title":"Improving Generalizability in Implicitly Abusive Language Detection with Concept Activation Vectors","year":2022,"lang":"en","type":"article","venue":"Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","topic":"Hate Speech and Cyberbullying Detection","field":"Computer Science","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada","funders":"","keywords":"Generalizability theory; Computer science; Robustness (evolution); Interpretability; Artificial intelligence; Moderation; Machine learning; Natural language processing; Metric (unit); Psychology","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":[0.001087185,0.0001502916,0.0002111844,0.0001152818,0.0005670655,0.00006072899,0.0006754858,0.00006506597,0.000001934768],"category_scores_gemma":[0.00406424,0.0001213549,0.0001428002,0.0008028617,0.00004934303,0.000124211,0.0003047006,0.000267517,2.647655e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006117028,"about_ca_system_score_gemma":0.000132866,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003706548,"about_ca_topic_score_gemma":0.00003444079,"domain_scores_codex":[0.9981428,0.0000585771,0.0004537935,0.0003197626,0.0007711988,0.0002539003],"domain_scores_gemma":[0.996874,0.0003246549,0.001329425,0.000136396,0.001302168,0.00003338899],"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":[0.0004178476,0.0004412454,0.2269587,0.000424487,0.0002629977,6.345239e-7,0.01930559,0.6508276,0.0774079,0.01697499,0.0004457559,0.006532231],"study_design_scores_gemma":[0.003919885,0.001191264,0.3284417,0.0003277796,0.0002084586,0.00001672677,0.00600937,0.4175321,0.2298997,0.009113396,0.002284722,0.001054865],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9917145,0.00002160325,0.005323723,0.0003829543,0.0009079913,0.0008324391,0.00008265046,0.00007727234,0.0006569048],"genre_scores_gemma":[0.9955375,3.144494e-7,0.003935311,0.0001059481,0.0001442763,0.0000549189,0.000007461156,0.0000169797,0.0001972434],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2332955,"threshold_uncertainty_score":0.4948711,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.003906633246131122,"score_gpt":0.2045248411760247,"score_spread":0.2006182079298936,"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."}}