Elamur Research⌕
☰
home / papers / 2026 / PatentMind: A Multi-Aspect Reasoning Graph for Patent Similarity Evaluation

PatentMind: A Multi-Aspect Reasoning Graph for Patent Similarity Evaluation

Paper · 2026 · Yongmin Yoo, Qiongkai Xu and Longbing Cao

Summary

Judging how similar two patents are matters for prior art search, novelty checks and infringement risk. Keyword methods and dense embeddings handle this poorly: they miss domain nuance and give scores that are hard to interpret. The authors present PatentMind, an LLM-driven framework built as a Multi-Aspect Reasoning Graph (MARG). It reads a patent's title, abstract and claims and extracts three views: technical features, application domain and claim scope. It then asks the LLM to score each view separately on a 0–1 scale. A context-aware weighting step combines these scores using IPC codes and the full pair context, without any fine-tuning. For evaluation, the authors built PatentSimBench, a public set of 500 expert-annotated patent pairs with rationales. Run on GPT-4o-mini, PatentMind agrees closely with expert judgments (Pearson r = 0.938). It beats embedding-based, patent-specific and prompting baselines while producing per-dimension scores that people can inspect. This makes it a promising, interpretable aid for patent analysis, though the authors advise against using it as the sole basis for decisions.

Key points

  • The combination step has four LLM reasoning stages:
    • Domain Relationship Analysis, which uses IPC codes
    • Information Distribution Analysis
    • Dimension Relevance Assessment
    • Cross-validation Reasoning
  • PatentSimBench contains 500 USPTO English patent pairs with expert labels; the authors describe it as the first benchmark of its kind.
  • Accuracy with GPT-4o-mini: r = 0.938, MAE = 0.0918, mean residual −0.0197.
    • Very similar pairs tend to be scored too low.
    • Clearly dissimilar pairs tend to be scored too high.
  • Errors are largest in terminology-heavy fields such as metallurgy and biotech. One example is failing to treat "leaching" and "extraction" as synonyms.
  • Cost: about 8 API calls, over 13,000 tokens and 20–30 seconds per pair. This suits re-ranking small candidate sets, not large-scale retrieval.

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

🎓Macquarie UniversitySydney · Australia44 works · 5,778 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksBusiness, Management and Accounting · 40 worksManagement of Technology and Innovation · 6 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Yongmin Yoo, Qiongkai Xu and Longbing Cao. 2026. PatentMind: A Multi-Aspect Reasoning Graph for Patent Similarity Evaluation. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.735

@inproceedings{yoo2026patentmind,
  title = {{PatentMind: A Multi-Aspect Reasoning Graph for Patent Similarity Evaluation}},
  author = {Yongmin Yoo, Qiongkai Xu and Longbing Cao},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.735},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.735}
}

References

28 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sam Arts, Bruno Cassiman, and Juan Carlos Gomez. 2018. Text matching to measure patent similarity. Strategic Management Journal, 39(1):62–84. find
  2. Grazia Sveva Ascione and Valerio Sterzi. 2024. A comparative analysis of embedding models for patent similarity. arXiv preprint arXiv:2403.16630. — [arXiv:2403.16630](https://arxiv.org/abs/2403.16630) find
  3. Grigor Aslanyan and Ian Wetherbee. 2022. Patents phrase to phrase semantic matching dataset. arXiv preprint arXiv:2208.01171. — [arXiv:2208.01171](https://arxiv.org/abs/2208.01171) find
  4. Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. Scibert: A pretrained language model for scientific text. arXiv preprint arXiv:1903.10676.
  5. Mihai Lupu and Allan Hanbury. 2013. Patent retrieval. Foundations and Trends in Information Retrieval, 7(1):1–97. find
  6. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. In Advances in Neural Information Processing Systems.
  7. Marko Miric, Nan Jia, and Kevin G Huang. 2023. Using supervised machine learning for large-scale classification in management research: The case for identifying artificial intelligence patents. Strategic Management Journal, 44(2):491–519. find
  8. Silvia Casola and Alberto Lavelli. 2021. Summarization, simplification, and generation: The case of patents. arXiv preprint arXiv:2104.14860. — [arXiv:2104.14860](https://arxiv.org/abs/2104.14860) find
  9. Eva D’hondt, Suzan Verberne, Cornelis Koster, and Lou Boves. 2013. Text representations for patent classification. Computational Linguistics, 39(3):755– 775. find
  10. Sijie Feng. 2020. The proximity of ideas: An analysis of patent text using machine learning. PLOS One, 15(7):e0234880. find
  11. Daniel S. Hain, Roman Jurowetzki, Tobias Buchmann, and Patrick Wolf. 2022. A text-embedding-based approach to measuring patent-to-patent technological similarity. Technological Forecasting and Social Change, 177:121559. find
  12. Lea Helmers, Franziska Horn, Franziska Biegler, Tim Oppermann, and Klaus-Robert Müller. 2019. Automating the search for a patent’s prior art with a full text similarity search. PLOS One, 14(3):e0212103. find
  13. Hayato Ikoma and Teruko Mitamura. 2025. Can ai examine novelty of patents?: Novelty evaluation based on the correspondence between patent claim and prior art. arXiv preprint arXiv:2502.06316. — [arXiv:2502.06316](https://arxiv.org/abs/2502.06316) find
  14. K. V. Indukuri, A. A. Ambekar, and A. Sureka. 2007. Similarity analysis of patent claims using natural language processing techniques. In Proceedings of the International Conference on Computational Intelligence and Multimedia Applications (ICCIMA’07), pages 169–175. IEEE. find
  15. Daeseong Jeon, Joon Mo Ahn, Juram Kim, and Changyong Lee. 2022. A doc2vec and local outlier factor approach to measuring the novelty of patents. Technological Forecasting and Social Change, 174:121294. find
  16. Lekang Jiang and Stephan Goetz. 2024. Natural language processing in patents: A survey. arXiv preprint arXiv:2403.04105. — [arXiv:2403.04105](https://arxiv.org/abs/2403.04105) find
  17. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. In Advances in Neural Information Processing Systems.
  18. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084.
  19. Yuen-Hsien Tseng, Chi-Jen Lin, and Yu-I Lin. 2007. Text mining techniques for patent analysis. Information Processing & Management, 43(5):1216–1247. find
  20. USPTO. 2024. Manual of Patent Examining Procedure, 9th edition, revision 01.2024 edition. United States Patent and Trademark Office. Accessed: 2025-03-22. find
  21. USPTO. 2025. Bulk data storage system (bdss) and open data portal (odp). https://bulkdata.uspto.gov/. BDSS retirement: April 11, 2025; ODP launched: February 12, 2025. ↗
  22. Meiyun Wang, Hiroki Sakaji, Hiroaki Higashitani, Mitsuhiro Iwadare, and Kiyoshi Izumi. 2023. Discovering new applications: Cross-domain exploration of patent documents using causal extraction and similarity analysis. World Patent Information, 74:102238. find
  23. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. 2022. Chain of thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems.
  24. WIPO. 1975. International patent classification (ipc). https://www.wipo.int/classifications/ipc/. IPC established by the Strasbourg Agreement, entered into force October 7, 1975. ↗
  25. Shunyu Yao et al. 2023. Tree of thoughts: Deliberate problem solving with large language models. arXiv preprint arXiv:2305.10601. find
  26. Yongmin Yoo, Cheonkam Jeong, Sanguk Gim, Junwon Lee, Zachary Schimke, and Deaho Seo. 2023. A novel patent similarity measurement methodology: Semantic distance and technological distance. arXiv preprint arXiv:2303.16767. — [arXiv:2303.16767](https://arxiv.org/abs/2303.16767) find
  27. Yongmin Yoo, Qiongkai Xu, and Longbing Cao. 2025. PatentScore: Multi-dimensional evaluation of LLM-generated patent claims. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 30727–30746, Suzhou, China. Association for Computational Linguistics. find
  28. Joon Lee and James Hsiang. 2020. Patentbert: Patent classification with fine-tuning a pre-trained bert model. World Patent Information, 62:101965. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yoo-2026-patentmind-multi-aspect-reasoning-graph/