ToxiREX is a new dataset of 128k Reddit comments in six languages with hierarchical annotations for implicit toxicity in conversational context based on an existing reasoning schema.
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , month = nov, year =
6 Pith papers cite this work, alongside 165 external citations. Polarity classification is still indexing.
6
Pith papers citing it
165
external citations · OpenAlex
representative citing papers
Interpreting harmful Discord messages requires integrating external knowledge and extended context, not just local message-level classification; LLMs leverage local context better than humans but still fail on coded language and community-specific references.
Cross-lingual analysis of 1.76M Singapore comments finds culturally specific hate targets but shared binding moral grammar and threat frames across languages.
ImpSH improves cross-domain generalization in implicit hate speech classification by aligning posts with implied statements and applying context-bounded semi-hard negative mining within a triplet learning setup.
CITA generates Chinese implicit toxicity samples that cause 69.48% average missed detection across seven tested detectors while preserving harmfulness, and the same data improves robustness when used to fine-tune a CITD defense model.
Censored LLMs achieve 69.0% strict accuracy in hate speech detection versus 64.1% for uncensored models and resist persona-based ideological influence better, but all exhibit overconfidence, irony failures, and group fairness disparities.
citing papers explorer
-
ToxiREX: A Dataset on Toxic REasoning in ConteXt
ToxiREX is a new dataset of 128k Reddit comments in six languages with hierarchical annotations for implicit toxicity in conversational context based on an existing reasoning schema.
-
Understanding Interpretation Difficulty in Harmful Online Communication: Insights from Cybercrime Communities
Interpreting harmful Discord messages requires integrating external knowledge and extended context, not just local message-level classification; LLMs leverage local context better than humans but still fail on coded language and community-specific references.
-
Cultural Targets, Structural Frames, Binding Morals: A Cross-Lingual Audit of Online Hate in Multicultural Singapore
Cross-lingual analysis of 1.76M Singapore comments finds culturally specific hate targets but shared binding moral grammar and threat frames across languages.
-
Aligning Implied Statements for Implicit Hate Speech Generalizability with Context-Bounded Semi-hard Negative Mining
ImpSH improves cross-domain generalization in implicit hate speech classification by aligning posts with implied statements and applying context-bounded semi-hard negative mining within a triplet learning setup.
-
Harder to Defend: Towards Chinese Toxicity Attacks via Implicit Enhancement and Obfuscation Rewriting
CITA generates Chinese implicit toxicity samples that cause 69.48% average missed detection across seven tested detectors while preserving harmfulness, and the same data improves robustness when used to fine-tune a CITD defense model.
-
Confident, Calibrated, or Complicit: Safety Alignment and Ideological Bias in LLM Hate Speech Detection
Censored LLMs achieve 69.0% strict accuracy in hate speech detection versus 64.1% for uncensored models and resist persona-based ideological influence better, but all exhibit overconfidence, irony failures, and group fairness disparities.