Elamur Research⌕
☰
home / papers / 2026 / RegNLI: Detecting Online Product Misbranding through Legal and Linguistic Alignment

RegNLI: Detecting Online Product Misbranding through Legal and Linguistic Alignment

Paper · 2026 · Diya Saha, Abhishek Bharadwaj Varanasi, Tirthankar Dasgupta et al.

Summary

This industry-track paper tackles automatic detection of misbranded health products, such as supplements, herbal remedies, cosmetics and drugs, whose marketing makes misleading claims prohibited by the U.S. FD&C Act. The authors argue that keyword matching and generic classifiers cannot link a claim to the legal provision it breaks. They therefore treat the task as inference between a product claim and regulatory text. They built FDA-Misbrand from FDA warning letters, using an LLM to extract product names, claims, violation explanations and cited sections. Experts checked part of the data, and packaging images were annotated with boxes marking the offending text. The RegNLI model has three components: a regulation-aware gate that steers claim representations toward compliance-relevant features, a contrastive loss that pulls claim–image embeddings toward the correct statute, and a binary misbranding classifier. Statutes are encoded as a hierarchical graph. The authors report gains over strong baselines in accuracy, F1 and statute alignment, along with interpretable attention on cues like quantifiers, guarantee verbs and hedges. The work matters for regulators who need predictions grounded in specific statutes.

Key points

  • FDA-Misbrand dataset:
    • Built from about 3,500 FDA warning letters (2015–2024) covering roughly 4,000 products.
    • Yields about 8,875 product–claim–regulation records and 27 regulatory acts.
    • Includes about 2,800 packaging images with violation-specific bounding boxes.
  • Extraction and validation:
    • Fields were extracted with GPT-4o mini.
    • Inter-annotator agreement on a 500-instance sample reached Cohen's κ = 0.82.
  • Statute encoding: Statutes were parsed with LegalBERT plus a Title → Chapter → Section → Clause hierarchy, forming a regulation graph in which clauses inherit context from parent sections (e.g., Section 502(a)).
  • Training objective: A temperature-scaled cosine contrastive alignment loss against negative regulations is combined with binary cross-entropy for classification. The exact scores fall in a portion of the text that was not provided.
  • Known weaknesses:
    • Vague claims such as "supports vitality" and hedged wording such as "may help" are hard to classify.
    • Statutes with conditional clauses are difficult to parse.
    • Claims bundling several assertions can be only partly aligned.
    • The model is sensitive to adversarial paraphrasing.
    • It covers the US only, with no multilingual or cross-jurisdiction support.

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

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksInformation Systems · 1,730 worksSociology and Political Science · 626 works

🧭 Topics 1Research subjects and languages the work is about.Legal AI · 150 works

Cite

Cite as Diya Saha, Abhishek Bharadwaj Varanasi, Tirthankar Dasgupta and Manjira Sinha. 2026. RegNLI: Detecting Online Product Misbranding through Legal and Linguistic Alignment. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-industry.31

@inproceedings{saha2026regnli,
  title = {{RegNLI: Detecting Online Product Misbranding through Legal and Linguistic Alignment}},
  author = {Diya Saha, Abhishek Bharadwaj Varanasi, Tirthankar Dasgupta and Manjira Sinha},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-industry.31},
  url = {https://doi.org/10.18653/v1/2026.eacl-industry.31}
}

References

18 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Manish Adawadkar. 2025. Ai-powered detection of deceptive product feedback: A review of methods, models, and future directions. International Journal of Trend in Scientific Research and Development (IJTSRD). find
  2. Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakasiotis, Nikolaos Aletras, and Ion Androutsopoulos. 2020. Legal-bert: The muppets straight out of law school. arXiv preprint arXiv:2010.02559. ↗
  3. Sudeshna Jana, Manjira Sinha, and Tirthankar Dasgupta. 2024. Force: A benchmark dataset for foodborne disease outbreak and recall event extraction from news. In Proceedings of The 9th Social Media Mining for Health Research and Applications (SMM4H 2024) Workshop and Shared Tasks, pages 163–169. find
  4. Kutubuddin Sayyad Liyakat Kazi, Sunita Sunil Shinde, Priya Mangesh Nerkar, Sultanabanu SayyadLiyakat Kazi, and Vahidabegam SayyadLiyakat Kazi. 2025. Machine learning for brand protection: A review of a proactive defense mechanism. Avoiding Ad Fraud and Supporting Brand Safety: Programmatic Advertising Solutions, pages 175–220. find
  5. Hariharan Pappil Kothandapani. 2025. Ai-driven regulatory compliance: Transforming financial oversight through large language models and automation. Emerging Science Research, 12(1):12–24. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33:9459– 9474. ↗
  6. Haitao Li, Junjie Chen, Jingli Yang, Qingyao Ai, Wei Jia, Youfeng Liu, Kai Lin, Yueyue Wu, Guozhi Yuan, Yiran Hu, and 1 others. 2025. Legalagentbench: Evaluating llm agents in legal domain. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2322–2344.
  7. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning, pages 19730–19742. PMLR. ↗
  8. Yam B Limbu, Christopher McKinley, and Valerio Tem-perini. 2019. A longitudinal examination of fda warning and untitled letters issued to pharmaceutical companies for violations in drug promotion standards. Journal of Consumer Affairs, 53(1):3–23. ↗
  9. Donna M Lisi. 2025. Ai for detecting and preventing adverse drug events. US Pharm, 50(2):32–37. find
  10. Moyang Liu, Kaiying Yan, Yukun Liu, Ruibo Fu, Zhengqi Wen, Xuefei Liu, and Chenxing Li. 2024. Misd-moe: A multimodal misinformation detection framework with adaptive feature selection. In NeurIPS Efficient Natural Language and Speech Processing Workshop. PMLR, pages 114–122. find
  11. Mason Marks. 2021. Automating fda regulation. Duke LJ, 71:1207. ↗
  12. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, and 1 others. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748–8763. PmLR.
  13. Marco Luca Sbodio, Vanessa López, Thanh Lam Hoang, Theodora Brisimi, Gabriele Picco, Inge Vejsbjerg, Valentina Rho, Pol Mac Aonghusa, Morten Kristiansen, and John Segrave-Daly. 2024. Collaborative artificial intelligence system for investigation of healthcare claims compliance. Scientific reports, 14(1):11884. ↗
  14. Gautam Kishore Shahi. 2025. Multimodal misinformation detection using early fusion of linguistic, visual, and social features. In Companion Publication of the 17th ACM Web Science Conference 2025, pages 11–18. ↗
  15. Bhupinder Singh. 2025. Sidestepping ad fraud through interfaces of artificial intelligence machine learning: Deep dive into financial fraud auxiliary brand safety. In Avoiding Ad Fraud and Supporting Brand Safety: Programmatic Advertising Solutions, pages 329–352. IGI Global Scientific Publishing. ↗
  16. Yoshiyasu Takefuji. 2025. Ai-driven analysis of drug marketing efficiency: Unveiling fda approval to market release dynamics. The AAPS Journal, 27(2):48. ↗
  17. Qingzheng Xu, Heming Du, Szymon Łukasik, Tianqing Zhu, Sen Wang, and Xin Yu. 2025. Mdam3: A misinformation detection and analysis framework for multitype multimodal media. In Proceedings of the ACM on Web Conference 2025, pages 5285–5296. find
  18. Michihiro Yasunaga, Jure Leskovec, and Percy Liang. 2022. Linkbert: Pretraining language models with document links. arXiv preprint arXiv:2203.15827. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/saha-2026-regnli-detecting-online-product-misbranding/