Elamur Research⌕
☰
home / papers / 2026 / An NLP Framework for Analyzing Corporate Strategic Behavior in the Opioid Industry Documents Archive

An NLP Framework for Analyzing Corporate Strategic Behavior in the Opioid Industry Documents Archive

Paper · 2026 · Duy Dang Phu and Thìn Đặng Văn

Summary

The Opioid Industry Documents Archive (OIDA) holds a large collection of internal corporate records released through opioid litigation. Its OCR-derived text is noisy and its metadata is sparse, so it has rarely been used for systematic computational study of corporate strategy. The authors build a multi-stage pipeline around presentation documents from Insys Therapeutics and Mallinckrodt Pharmaceuticals:

  • The OCR text is cleaned and split into overlapping chunks.
  • Chunks are filtered with a keyword list, then with a KNN classifier trained on LLM-labelled samples.
  • The remaining chunks are clustered with BERTopic.
  • An LLM assigns each topic to one of three strategy categories and writes an interpretation that must stay tied to the source excerpts.

The two firms show different profiles. Insys topics are dominated by sales and promotion aimed at individual prescribers, concentrated in 2011–2015. Mallinckrodt shows more emphasis on market access, payer dynamics and regulatory risk. The work offers a scalable, reproducible way to explore litigation archives for public-health and accountability research. However, the LLM outputs have not yet been checked by domain experts.

Key points

  • Preprocessing and models: presentations are cut into 300-word chunks with a 50-word overlap. The same embedding model, avsolatorio/GIST-Embedding-v0, is used for both filtering and BERTopic. Topic keywords come from the KeyBERTInspired representation.
  • Relevance filtering: an LLM using chain-of-thought prompting labelled 150 random segments per company. A KNN classifier (k = 5, selected by F1) then spread these labels across the full dataset.
  • Strategy categories: Sales Expansion & Promotion, Influence & Narrative Management, and Regulatory Risk Management & Evasion. A single topic can fall into more than one category.
  • Topic counts:
    • Insys: 51 topics in total, of which 24 are sales, 5 influence and 0 regulatory risk.
    • Mallinckrodt: 25 topics in total, of which 9 are sales, 1 influence and 5 regulatory risk.
    • Insys activity overlaps with the period after the FDA approved Subsys in January 2012.
  • Limitations noted by the authors:
    • Only presentation documents are analysed.
    • The archive's temporal coverage is uneven, especially before 2010.
    • Raw document counts are used to measure intensity.
    • Prompts are tuned for precision, so implicit strategies may be missed. Finding no legal-avoidance strategies therefore does not prove they were absent.

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

🎓Vietnam National University Ho Chi Minh CityHo Chi Minh City · Vietnam21 works · 14 citations🎓University of Information TechnologyHo Chi Minh City · Vietnam13 works · 7 citations

 Fields, topics and concepts

📚 Shelves 8The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksGeneral Social Sciences · 227 worksMedicine · 241 worksPublic Health, Environmental and Occupational Health · 10 works

Cite

Cite as Duy Dang Phu and Thìn Đặng Văn. 2026. An NLP Framework for Analyzing Corporate Strategic Behavior in the Opioid Industry Documents Archive. Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.nlpcss-1.7

@inproceedings{phu2026nlp,
  title = {{An NLP Framework for Analyzing Corporate Strategic Behavior in the Opioid Industry Documents Archive}},
  author = {Duy Dang Phu and Thìn DJặng Văn},
  booktitle = {Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science},
  year = {2026},
  doi = {10.18653/v1/2026.nlpcss-1.7},
  url = {https://doi.org/10.18653/v1/2026.nlpcss-1.7}
}

References

13 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ibrahim Al Azher, Venkata Devesh Reddy Seethi, Akhil Pandey Akella, and Hamed Alhoori. 2025. Limtopic: Llm-based topic modeling and text summarization for analyzing scientific articles limitations. In Proceedings of the 24th ACM/IEEE Joint Conference on Digital Libraries, New York, NY, USA. Associat… find
  2. Alessio Faccia, Julie McDonald, and Babu George. 2024. Nlp sentiment analysis and accounting transparency: A new era of financial record keeping. Computers, 13(1). find
  3. Maarten Grootendorst. 2022. Bertopic: Neural topic modeling with a class-based tf-idf procedure. arXiv preprint arXiv:2203.05794.
  4. Hyewon Kang and Jinho Kim. 2022. Analyzing and visualizing text information in corporate sustainability reports using natural language processing methods. Applied Sciences, 12(11). find
  5. Opioid Industry Documents Archive. n.d. Timeline of the opioid crisis. https://timeline.oida-resources.jhu.edu/. Accessed: 2 March 2026. ↗
  6. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12:282… find
  7. Jantima Polpinij, Manasawee Kaenampornpan, Christopher S. G. Khoo, Wei-Ning Cheng, and Bancha Lu-aphol. 2026. A multi-stage nlp framework for knowledge discovery from crop disease research literature. Mathematics, 14(2). find
  8. Xuan Shen, Brian Wingenroth, Zichao Wang, Jason Kuen, Wanrong Zhu, Ruiyi Zhang, Yiwei Wang, Lichun Ma, Anqi Liu, Hongfu Liu, and 1 others. 2025. Oida-qa: A multimodal benchmark for analyzing the opioid industry documents archive. arXiv preprint arXiv:2511.09914. — [arXiv:2511.09914](https://arxiv.org/abs/2511.09914) find
  9. Aivin V. Solatorio. 2024. Gistembed: Guided in-sample selection of training negatives for text embedding fine-tuning. arXiv preprint arXiv:2402.16829. — [arXiv:2402.16829](https://arxiv.org/abs/2402.16829) find
  10. University of California, San Francisco and Johns Hopkins University. n.d. Opioid industry documents archive (oida). https://www.industrydocuments.ucsf.edu/opioids/. Accessed: 6 February 2026. ↗
  11. U.S. Department of Justice. 2019. Founder and four executives of insys therapeutics convicted of racketeering conspiracy. Accessed: 4 March 2026. find
  12. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  13. Xiaohao Yang, He Zhao, Dinh Phung, Wray Buntine, and Lan Du. 2025. LLM reading tea leaves: Automatically evaluating topic models with large language models. Transactions of the Association for Computational Linguistics, 13:357–375. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/phu-2026-nlp-framework-analyzing-corporate-strategic/