Elamur Research⌕
☰
home / papers / 2026 / DPR@DravidianLangTech 2026: Transformer-Based Abusive Content Detection for Tamil Text Targeting Women on Social Media

DPR@DravidianLangTech 2026: Transformer-Based Abusive Content Detection for Tamil Text Targeting Women on Social Media

Paper · 2026 · Diya Prakash, Praveen Kumar S, R Ranjith Kumar et al.

Summary

The paper addresses the automatic detection of abusive, women-targeted content in Tamil social media comments. This task is hard because of Tamil's rich morphology, dialectal variety, transliteration, code-mixing with English, and sarcasm. These features make simple keyword lists unreliable.

Working within the DravidianLangTech 2026 shared task, the authors framed the problem as binary classification of YouTube comments. They fine-tuned five pretrained encoders on the official dataset, each with a classification head on the [CLS] representation: mBERT, MuRIL, XLM-RoBERTa, IndicBERT, and Tamil-BERT. They compared the models and also examined how multilingual versus Tamil-specific pretraining and WordPiece versus SentencePiece tokenization affect results.

Tamil-BERT performed best, reaching 80.72% accuracy and placing 5th on the shared-task leaderboard. MuRIL came close behind. A qualitative error analysis showed that multilingual models tend to misread colloquial slang, sarcasm, and emotional tone. The study suggests that language-specific or Indic-focused pretraining helps content moderation in low-resource languages. Code and fine-tuned models are publicly released on GitHub.

Key points

  • Dataset: the DravidianLangTech 2026 Tamil YouTube comments, labeled Abusive against Women or Non-Abusive.
    • Training split: 3,652 comments (1,769 abusive, 1,883 non-abusive).
    • Test split: 913 comments (441 abusive, 472 non-abusive).
  • Test accuracy and Macro F1 by model:
    • Tamil-BERT: accuracy 0.8072, Macro F1 0.8072.
    • MuRIL: accuracy 0.7919, Macro F1 0.7889.
    • mBERT: accuracy 0.7711, Macro F1 0.7848.
    • XLM-RoBERTa: accuracy 0.7338, Macro F1 0.7338.
    • IndicBERT: accuracy 0.7049, Macro F1 0.7313.
  • Fine-tuning setup:
    • AdamW optimizer with cross-entropy loss.
    • Maximum sequence length 128, batch size 32, learning rate 2e-5, 3 epochs.
    • Minimal preprocessing, so that slang and transliteration are kept.
  • Authors' explanations for the ranking:
    • MuRIL's training data includes transliterated text, which helps with "Tanglish".
    • IndicBERT's lightweight ALBERT architecture likely limits its capacity.
  • Future directions: domain-adaptive pretraining on Tamil social media text, data augmentation, and ensemble or hybrid architectures.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Indian Institute of Information Technology KottayamPalai · India2 works · 1 citation

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksPsychology · 1,552 worksSocial Psychology · 306 works

Cite

Cite as Diya Prakash, Praveen Kumar S, R Ranjith Kumar, Balasubramanian Palani, Jobin Jose and Siranjeevi Rajamanickam. 2026. DPR@DravidianLangTech 2026: Transformer-Based Abusive Content Detection for Tamil Text Targeting Women on Social Media. Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.dravidianlangtech-1.35

@inproceedings{prakash2026dpr,
  title = {{DPR@DravidianLangTech 2026: Transformer-Based Abusive Content Detection for Tamil Text Targeting Women on Social Media}},
  author = {Diya Prakash, Praveen Kumar S, R Ranjith Kumar, Balasubramanian Palani, Jobin Jose and Siranjeevi Rajamanickam},
  booktitle = {Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages},
  year = {2026},
  doi = {10.18653/v1/2026.dravidianlangtech-1.35},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.35}
}

References

14 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. A. Abishek and T. C. Govindapillai. 2024. Detecting tamil textual threats in social media using artificial intelligence. International Journal of Novel Research and Development (IJNRD), 9(3):26–33. find
  2. M. Alam and S. Rahman. 2025. Cuet_ignite@dravidianlangtech 2025: Detection of abusive comments in tamil text using transformer models. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 411–418. Association for Computational Linguistics. find
  3. S. Bhattacharyya, P. Gupta, and M. Lowe. 2025. Lexilogic@dravidianlangtech 2025: Detecting misogynistic memes and abusive tamil and malayalam text targeting women on social media. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 112–119… find
  4. U. Hegde and A. Hande. 2025. Team_strikers@dravidianlangtech 2025: Misogyny meme detection in tamil using multimodal deep learning. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 605–612. Association for Computational Linguistics. find
  5. A. Kalaivani and D. Thenmozhi. 2024. Composite feature fusion for improved offensive language detection in tamil social media using mha-lstm. International Journal of Machine Learning and Cybernetics, 15:1– 15. find
  6. Kumaresan and S. Karnati. 2025. Kec_ai_vss_run2@dravidianlangtech 2025: Abusive tamil and malayalam text targeting women on social media. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 315–322. Association for Computational Linguistic… find
  7. A. Mallik, R. Dhar, U. Das, M. A. Labib, S. Rahman, and H. Murad. 2025. Cuet-823@dravidianlangtech 2025: Shared task on multimodal misogyny meme detection in tamil language. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 401– 408. Ass… find
  8. J. Mohan, S. R. Mekapati, B. Premjith, G. J. Lal, and B. R. Chakravarthi. 2025. A multimodal approach for hate and offensive content detection in tamil: From corpus creation to model development. ACM Transactions on Asian and Low-Resource Language Information Processing, 24(3):1–24. find
  9. J. Pokrywka and K. Jassem. 2024. Evaluating transformer models for hate speech detection in tamil. In Proceedings of the Fourth Workshop on Language Technology for Equality, Diversity, Inclusion, pages 196–199. Association for Computational Linguistics. find
  10. R. Ponnusamy, B. Sivagnanam, and B. R. Chakravarthi. 2024. From laughter to inequality: Annotated dataset for misogyny detection in tamil and malayalam memes. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING), page… find
  11. S. Rajiakodi, B. R. Chakravarthi, S. P. M. Chinnan, R. Priyadharshini, J. Rajameenakshi, and 1 others. 2025. Findings of the shared task on abusive tamil and malayalam text targeting women on social media: Dravidianlangtech@naacl 2025. In Proceedings of the Fifth Workshop on Speech, Vision, and Lang… find
  12. Sangeetham and M. Bedi. 2025. Jas@dravidianlangtech 2025: Abusive tamil text targeting women on social media. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 210–216. Association for Computational Linguistics. find
  13. M. Subramanian, K. S. Vadivel, and R. Sowmya. 2022. Detecting offensive tamil texts using machine learning and multilingual transformer models. In Proceedings of the 2022 International Conference on Smart Technologies and Systems for Next Generation Computing (ICSTSN), pages 1–6. IEEE. find
  14. Thayasivam and Y. S. Wijesuriya. 2025. Gs_dravidianlangtech@2025: Women targeted abusive texts detection on social media. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 520–527. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/prakash-2026-dpr-dravidianlangtech-2026-transformer-based/