Elamur Research⌕
☰
home / papers / 2026 / STEM TOX: From Collaborative Tags to Fine-Grained Toxic Meme Detection via Entropy-Guided Multi-Task Learning

STEM TOX: From Collaborative Tags to Fine-Grained Toxic Meme Detection via Entropy-Guided Multi-Task Learning

Paper · 2026 · Subhankar Swain, Naquee Rizwan, Vishwa Gangadhar S et al.

Summary

The paper addresses toxic meme moderation, which is held back by scarce real-world data and by vision-language models that miss contextual cues. The authors introduce ToxicTags, a dataset of 6,300 memes crawled from three imgflip streams (Dark_humour, MEMES_OVERLOAD, politics). The memes were not filtered by target or event. Each meme is kept with its post metadata, including the title, OCR text and the user-assigned collaborative tags. Annotation was done in two stages: first toxic vs. normal, then toxic items split into hateful, dangerous or offensive.

They also propose STEMTOX, a multi-task framework built on LoRA-adapted VLMs (PaliGemma, LLaVA). It produces tags for a meme while also classifying it. For classification, it selects the lowest-entropy token embedding across layers as the representation.

STEMTOX outperforms competing classifiers and tag generators, and it also generalizes to the FHM and MAMI benchmarks. The work shows that community tags supply useful context for detecting harmful multimodal content. It also offers an openly released resource for training moderation systems.

Key points

  • Dataset collection: ToxicTags was drawn from 37,072 crawled memes. Only posts with at least two comments were kept, giving a final set of 6,300 labelled memes. The label scheme has four classes: hateful, dangerous, offensive and normal. The code and the Hugging Face dataset are public.
  • Classification results: STEMTOX reaches 72.55% macro-F1 on Stage I and 64.66% on Stage II. These are gains of 2.07 and 3.12 points over the strongest baseline.
  • Tag generation: With the PaliGemma backbone, tag generation scores 0.4872 chrF, 0.265 METEOR and 0.626 semantic similarity. This beats Tag2Text, RAM and RAM++ by about 0.28, 0.18 and 0.29 respectively. PaliGemma also outperforms LLaVA on tag generation.
  • Generalization: On FHM, STEMTOX improves accuracy and macro-F1 by about 3.3 points over prior methods. On MAMI, the gains are roughly 6.3–6.5 points.
  • Error analysis:
    • LLaVA tends to fail on memes tagged with death, mental-health or religious topics.
    • PaliGemma struggles with slang and sarcasm.
    • Both models show the highest uncertainty, measured with maximum softmax probability, on the offensive class.

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 Technology KharagpurKharagpur · India24 works · 714 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksComputer Vision and Pattern Recognition · 2,869 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 1Research subjects and languages the work is about.Bias, fairness & safety · 517 works

Cite

Cite as Subhankar Swain, Naquee Rizwan, Vishwa Gangadhar S, Nayandeep Deb and Animesh Mukherjee. 2026. STEM TOX: From Collaborative Tags to Fine-Grained Toxic Meme Detection via Entropy-Guided Multi-Task Learning. Transactions of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/tacl.a.801

@article{swain2026stem,
  title = {{STEM TOX: From Collaborative Tags to Fine-Grained Toxic Meme Detection via Entropy-Guided Multi-Task Learning}},
  author = {Subhankar Swain, Naquee Rizwan, Vishwa Gangadhar S, Nayandeep Deb and Animesh Mukherjee},
  journal = {Transactions of the Association for Computational Linguistics},
  year = {2026},
  doi = {10.1162/tacl.a.801},
  url = {https://doi.org/10.1162/tacl.a.801}
}

References

46 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Carlos Arcila Calderón, Patricia Sánchez Holgado, Jesús Gómez, Marcos Barbosa, Haodong Qi, Alberto Matilla, Pilar Amado, Alejandro Guzmán, Daniel López-Matías, and Tomás Fernández-Villazala. 2024. From online hate speech to offline hate crime: the role of inflammatory language in forecasting violenc… find
  2. Aashish Bhandari, Siddhant B Shah, Surendrabikram Thapa, Usman Naseem, and Mehwish Nasim. 2023. Crisishatemm: Multimodal analysis of directed and undirected hate speech in text-embedded images from russia-ukraine conflict. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco… find
  3. Fariha Tahosin Boishakhi, Ponkoj Chandra Shill, and Md. Golam Rabiul Alam. 2021. Multimodal hate speech detection using machine learning. In 2021 IEEE International Conference on Big Data (Big Data), page 4496–4499. IEEE. find
  4. Rui Cao, Ming Shan Hee, Adriel Kuek, Wen-Haw Chong, Roy Ka-Wei Lee, and Jing Jiang. 2023. Pro-cap: Leveraging a frozen vision-language model for hateful meme detection. In Proceedings of the 31st ACM international conference on multimedia, pages 5244–5252. find
  5. Rui Cao, Roy Ka-Wei Lee, Wen-Haw Chong, and Jing Jiang. 2022. Prompting for multimodal hateful meme classification. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 321–332, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics. find
  6. Rui Cao, Roy Ka-Wei Lee, and Jing Jiang. 2024a. Modularized networks for few-shot hateful meme detection. In Proceedings of the ACM Web Conference 2024, WWW ’24, page 4575–4584, New York, NY, USA. Association for Computing Machinery. find
  7. Rui Cao, Roy Ka-Wei Lee, and Jing Jiang. 2024b. Modularized networks for few-shot hateful meme detection. find
  8. Guanqi Chen, Lei Yang, Guanhua Chen, and Jia Pan. 2023. Evaluating explanation methods for vision-and-language navigation. find
  9. Haoran Chen, Junyan Lin, Xinghao Chen, Yue Fan, Jianfeng Dong, Xin Jin, Hui Su, Jinlan Fu, and Xiaoyu Shen. 2025. Multimodal language models see better when they look shallower. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 6677–6695, Suzhou, China.… find
  10. Yi Dai, Hao Lang, Kaisheng Zeng, Fei Huang, and Yongbin Li. 2023. Exploring large language models for multi-modal out-of-distribution detection. arXiv preprint arXiv:2310.08027. find
  11. Mithun Das and Animesh Mukherjee. 2023a. Banglaabusememe: A dataset for bengali abusive meme classification. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 15498– 15512. find
  12. Mithun Das and Animesh Mukherjee. 2023b. Transfer learning for multilingual abusive meme detection. In Proceedings of the 15th ACM Web Science Conference 2023, pages 245–250. find
  13. Thomas Davidson, Dana Warmsley, Michael Macy, and Ingmar Weber. 2017. Automated hate speech detection and the problem of offensive language. Proceedings of the International AAAI Conference on Web and Social Media, 11(1):512–515.
  14. Elisabetta Fersini, Francesca Gasparini, Giulia Rizzi, Aurora Saibene, Berta Chulvi, Paolo Rosso, Alyssa Lees, and Jeffrey Sorensen. 2022. Semeval-2022 task 5: Multimedia automatic misogyny identification. In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages…
  15. Zoe Wanying He, Sean Trott, and Meenakshi Khosla. 2025. Seeing through words, speaking through pixels: Deep representational alignment between vision and language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 35645–35660. find
  16. Ming Shan Hee, Aditi Kumaresan, and Roy Ka-Wei Lee. 2024. Bridging modalities: Enhancing cross-modality hate speech detection with few-shot in-context learning. arXiv preprint arXiv:2410.05600. find
  17. Hongzhan Lin, Ziyang Luo, Bo Wang, Ruichao Yang, and Jing Ma. 2026. Goat-bench: Safety insights to large multimodal models through meme-based social abuse. ACM Trans. Intell. Syst. Technol., 17(4). find
  18. Dan Hendrycks and Kevin Gimpel. 2017. A baseline for detecting misclassified and out-of-distribution examples in neural networks. In Proceedings of the International Conference on Learning Representations (ICLR). find
  19. Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2024. Improved baselines with visual instruction tuning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 26296– 26306.
  20. Xinyu Huang, Yi-Jie Huang, Youcai Zhang, Weiwei Tian, Rui Feng, Yuejie Zhang, Yanchun Xie, Yaqian Li, and Lei Zhang. 2025. Open-set image tagging with multi-grained text supervision. In Proceedings of the 33rd ACM International Conference on Multimedia, MM ’25, page 4117–4126, New York, NY, USA. Ass… find
  21. Xinyu Huang, Youcai Zhang, Jinyu Ma, Weiwei Tian, Rui Feng, Yuejie Zhang, Yaqian Li, Yandong Guo, and Lei Zhang. 2024. Tag2text: Guiding vision-language model via image tagging. In ICLR. find
  22. Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Pratik Ringshia, and Davide Testuggine. 2020. The hateful memes challenge: Detecting hate speech in multimodal memes. Advances in neural information processing systems, 33:2611– 2624.
  23. Gary King and Langche Zeng. 2001. Logistic regression in rare events data. Political Analysis, 9(2):137–163. find
  24. Girish A. Koushik, Helen Treharne, and Diptesh Kanojia. 2026. Tandem: Temporal-aware neural detection for multimodal hate speech. find
  25. Gitanjali Kumari, Dibyanayan Bandyopadhyay, Asif Ekbal, and Vinutha B. NarayanaMurthy. 2024. CM-off-meme: Code-mixed Hindi-English offensive meme detection with multi-task learning by leveraging contextual knowledge. In Proceedings of the 2024 Joint International Conference on Computational Linguist… find
  26. Junyu Lu, Bo Xu, Xiaokun Zhang, Haohao Zhu, Kaichun Wang, Liang Yang, and Hongfei Lin. 2025. Is having rationales enough? rethinking knowledge enhancement for multimodal hateful meme detection. In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information R… find
  27. Grace Luo, Trevor Darrell, and Amir Bar. 2024. Vision-language models create cross-modal task representations. arXiv preprint arXiv:2410.22330. find
  28. Atanu Mandal, Gargi Roy, Amit Barman, Indranil Dutta, and Sudip Kumar Naskar. 2024. Attentive fusion: A transformer-based approach to multimodal hate speech detection. find
  29. Binny Mathew, Punyajoy Saha, Seid Muhie Yimam, Chris Biemann, Pawan Goyal, and Animesh Mukherjee. 2021. Hatexplain: A benchmark dataset for explainable hate speech detection. Proceedings of the AAAI Conference on Artificial Intelligence, 35(17):14867–14875. find
  30. OpenAI. 2024. GPT-4o — openai.com. https://openai.com/index/hello-gpt-4 o/. ↗
  31. Nirmalendu Prakash, Han Wang, Nguyen Khoi Hoang, Ming Shan Hee, and Roy Ka-Wei Lee. 2023. Promptmtopic: Unsupervised multimodal topic modeling of memes using large language models. In Proceedings of the 31st ACM International Conference on Multimedia, MM ’23, page 621–631, New York, NY, USA. Associa… find
  32. Shraman Pramanick, Dimitar Dimitrov, Rituparna Mukherjee, Shivam Sharma, Md Shad Akhtar, Preslav Nakov, and Tanmoy Chakraborty. 2021a. Detecting harmful memes and their targets. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2783–2796.
  33. Shraman Pramanick, Shivam Sharma, Dimitar Dimitrov, Md Shad Akhtar, Preslav Nakov, and Tanmoy Chakraborty. 2021b. Momenta: A multimodal framework for detecting harmful memes and their targets. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 4439–4455.
  34. Yihan Ma1 Xinyue Shen1 Yiting Qu, Ning Yu2 Michael Backes, and Savvas Zannettou3 Yang Zhang. 2025. From meme to threat: On the hateful meme understanding and induced hateful content generation in open-source vision language models. In USENIX Security Symposium (USENIX Security). USENIX. find
  35. Naquee Rizwan, Paramananda Bhaskar, Mithun Das, Swadhin Satyaprakash Majhi, Punyajoy Saha, and Animesh Mukherjee. 2025a. Exploring the limits of zero shot vision language models for hate meme detection: The vulnerabilities and their interpretations. In Proceedings of the International AAAI Conferenc… find
  36. Naquee Rizwan, Nayandeep Deb, Sarthak Roy, Vishwajeet Singh Solanki, Kiran Garimella, and Animesh Mukherjee. 2025b. Toxicity begets toxicity: Unraveling conversational chains in political podcasts. In Proceedings of the 33rd ACM International Conference on Multimedia, MM ’25, page 11776–11784, New Y… find
  37. Paul Röttger, Bertie Vidgen, Dirk Hovy, and Janet Pierrehumbert. 2022. Two contrasting data annotation paradigms for subjective NLP tasks. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 175–190,…
  38. Sarthak Roy, Ashish Harshvardhan, Animesh Mukherjee, and Punyajoy Saha. 2023. Probing llms for hate speech detection: strengths and vulnerabilities. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 6116–6128. find
  39. Pratik Sachdeva, Renata Barreto, Geoff Bacon, Alexander Sahn, Claudia Von Vacano, and Chris Kennedy. 2022. The measuring hate speech corpus: Leveraging rasch measurement theory for data perspectivism. In Proceedings of the 1st Workshop on Perspectivist Approaches to NLP@ LREC2022, pages 83–94. find
  40. Surendrabikram Thapa, Farhan Ahmad Jafri, Kritesh Rauniyar, Mehwish Nasim, and Usman Naseem. 2024. Ruhate-mm: Identification of hate speech and targets using multimodal data from russia-ukraine crisis. In Companion Proceedings of the ACM Web Conference 2024, pages 1854–1863. find
  41. Surendrabikram Thapa, Aditya Shah, Farhan Ahmad Jafri, Usman Naseem, and Imran Razzak. 2022. A multi-modal dataset for hate speech detection on social media: Case-study of russia-ukraine conflict. In CASE 2022-5th Workshop on Challenges and Applications of Automated Extraction of Socio-Political Eve… find
  42. Sriram Vajapeyam. 2014. Understanding shannon’s entropy metric for information. find
  43. Wenxuan Wang, Xiaoyuan Liu, Kuiyi Gao, Jen-tse Huang, Youliang Yuan, Pinjia He, Shuai Wang, and Zhaopeng Tu. 2025. Can’t see the forest for the trees: Benchmarking multimodal safety awareness for multimodal LLMs. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguist… find
  44. Qiyao Xue, Yuchen Dou, Zheyuan Ryan Shi, Xiang Lorraine Li, and Wei Gao. 2026. Mmbert: Scaled mixture-of-experts multimodal bert for robust chinese hate speech detection under cloaking perturbations. Proceedings of the AAAI Conference on Artificial Intelligence, 40(40):34196–34204. find
  45. Akhila Yerukola, Saadia Gabriel, Nanyun Peng, and Maarten Sap. 2025. Mind the gesture: Evaluating ai sensitivity to culturally offensive non-verbal gestures. find
  46. System prompt for zero-shot and few-shot classification tasks with VLMs. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/swain-2026-stem-tox-collaborative-tags-fine/