mlsanityData to 7 Oct 2026

ToxiTrace: Gradient-Aligned Training for Explainable Chinese Toxicity Detection

arXiv 2604.12321 ↗First posted 14 Apr 2026cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

Existing Chinese toxic content detection methods mainly target sentence-level classification but often fail to provide readable and contiguous toxic evidence spans. We propose ToxiTrace, an explainability-oriented method for BERT-style encoders with three components: (1) CuSA, which refines encoder-derived saliency cues into fine-grained toxic spans with lightweight LLM guidance; (2) GCLoss, a gradient-constrained objective that concentrates token-level saliency on toxic evidence while suppressing irrelevant activations; and (3) ARCL, which constructs sample-specific contrastive reasoning pairs to sharpen the semantic boundary between toxic and non-toxic content. Experiments show that ToxiTrace improves classification accuracy and toxic span extraction while preserving efficient encoder-based inference and producing more coherent, human-readable explanations. We have released the model at https://huggingface.co/ArdLi/ToxiTrace.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 29 of its 46 references that are in the map. First posted on arXiv 14 Apr 2026.

Builds on

29 of 46 references.

References · 29

  1. 2017Attention Is All You Needcited by 59,932
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  4. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
  5. 2017Automated Hate Speech Detection and the Problem of Offensive Languagecited by 724
  6. 2020LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attentioncited by 326
  7. 2017Right for the Right Reasons: Training Differentiable Models by Constraining their Explanationscited by 293
  8. 2020Revisiting Pre-Trained Models for Chinese Natural Language Processingcited by 262
  9. 2020HateBERT: Retraining BERT for Abusive Language Detection in Englishcited by 206
  10. 2017Deep Learning for Hate Speech Detection in Tweetscited by 180
  11. 2022COLD: A Benchmark for Chinese Offensive Language Detectioncited by 74
  12. 2021"Will You Find These Shortcuts?" A Protocol for Evaluating the Faithfulness of Input Salience Methods for Text Classificationcited by 66
17 more
  1. 2021ChineseBERT: Chinese Pretraining Enhanced by Glyph and Pinyin Informationcited by 65
  2. 2021How can I choose an explainer? An Application-grounded Evaluation of Post-hoc Explanationscited by 58
  3. 2023Facilitating Fine-grained Detection of Chinese Toxic Language: Hierarchical Taxonomy, Resources, and Benchmarkscited by 45
  4. 2023Faithful Explanations of Black-box NLP Models Using LLM-generated Counterfactualscited by 44
  5. 2022Handling and Presenting Harmful Text in NLP Researchcited by 43
  6. 2021Evaluating Saliency Methods for Neural Language Modelscited by 37
  7. 2021Lexicon Enhanced Chinese Sequence Labeling Using BERT Adaptercited by 34
  8. 2021FBERT: A Neural Transformer for Identifying Offensive Contentcited by 30
  9. 2021Improving the Faithfulness of Attention-based Explanations with Task-specific Information for Text Classificationcited by 29
  10. 2024Evaluating the Performance of Large Language Models via Debatescited by 19
  11. 2021Enjoy the Salience: Towards Better Transformer-based Faithful Explanations with Word Saliencecited by 16
  12. 2025Exploring Multimodal Challenges in Toxic Chinese Detection: Taxonomy, Benchmark, and Findingscited by 10
  13. 2022Entity-to-Text based Data Augmentation for various Named Entity Recognition Taskscited by 6
  14. 2021Virtual Data Augmentation: A Robust and General Framework for Fine-tuning Pre-trained Modelscited by 5
  15. 2023Dynamic Top-k Estimation Consolidates Disagreement between Feature Attribution Methodscited by 4
  16. 2023Cyberbullying in Text Content Detection: An Analytical Reviewcited by 2
  17. 2025MMBERT: Scaled Mixture-of-Experts Multimodal BERT for Robust Chinese Hate Speech Detection under Cloaking Perturbationscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025SMARTER: A Data-efficient Framework to Improve Toxicity Detection with Explanation via Self-augmenting Large Language Models
  2. 2025Rethinking Toxicity Evaluation in Large Language Models: A Multi-Label Perspective
  3. 2025ToxiFrench: Benchmarking and Enhancing Language Models via CoT Fine-Tuning for French Toxicity Detection
  4. 2025Chinese Toxic Language Mitigation via Sentiment Polarity Consistent Rewrites
  5. 2025ShieldVLM: Safeguarding the Multimodal Implicit Toxicity via Deliberative Reasoning with LVLMs
  6. 2026ToxiREX: A Dataset on Toxic REasoning in ConteXt
  7. 2025Something Just Like TRuST : Toxicity Recognition of Span and Target
  8. 2024Just KIDDIN: Knowledge Infusion and Distillation for Detection of INdecent Memes