Elamur Research⌕
☰
home / papers / 2026 / Inventive Problem Solving with LLMs: A Benchmark for TRIZ Reasoning

Inventive Problem Solving with LLMs: A Benchmark for TRIZ Reasoning

Paper · 2026 · Zhu Wang and Brian Uzzi

Summary

LLMs are increasingly used as invention assistants. Real inventive work, however, depends on spotting the engineering trade-off in which improving one property harms another. TRIZ models such trade-offs as contradictions and links them to standard inventive principles. Earlier TRIZ–LLM studies were small, domain-specific case studies and seldom used patent text. The authors present TRIZBench, which combines 1,354 structured TRIZ cases from open papers and TRIZ web publications with 429 U.S. patents, some of which carry human-labeled TRIZ parameters and principles.

Cases were built in three steps:

  • experts annotated a seed set,
  • Gemini-2.5-pro detected relevant segments and extracted schema-based case objects,
  • automatic checks validated the schema and the evidence.

The benchmark asks models to do three things: identify the contradiction, select the applicable inventive principles, and produce grounded TRIZ reasoning backed by sentence-level patent evidence. Baseline results show three patterns:

  • Models detect that a contradiction exists more easily than they recover the correct improve–worsen pair.
  • Principle prediction improves when methods exploit TRIZ structure.
  • Semantic retrieval is essential for evidence-based justification.

The resource enables auditable evaluation of structured reasoning on patents.

Key points

  • The case corpus covers 590 source documents, 1,354 cases, 1,679 contradictions and 2,630 solutions across 10 major domains. 54.2% of cases carry TRIZ principles.
  • The patent side has 429 unique U.S. patents spanning 83 three-digit CPC codes. Of these, 257 have parameter labels and 304 have principle labels, and 148 cases link directly to patents.
  • Seed annotation by two experts covered 10 documents, 64 segments and 17 cases. PDFs were converted to markdown with marker, using Gemini-2.5-flash as the backend.
  • In grounded reasoning, TF-IDF retrieval helps little. PatentSBERTa aligns evidence best, gaining about 0.14 F1, and outperforms MiniLM and MPNet.
  • Better evidence F1 goes with higher principle Hit@3. Label accuracy alone can mislead: a correct parameter pair may still lack supporting evidence.

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 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 worksEngineering · 133 worksBusiness, Management and Accounting · 32 worksMechanical Engineering · 3 worksManagement of Technology and Innovation · 6 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Zhu Wang and Brian Uzzi. 2026. Inventive Problem Solving with LLMs: A Benchmark for TRIZ Reasoning. 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.1798

@inproceedings{wang2026inventive,
  title = {{Inventive Problem Solving with LLMs: A Benchmark for TRIZ Reasoning}},
  author = {Zhu Wang and Brian Uzzi},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1798},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1798}
}

References

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

  1. Amna Ali, Ali Tufail, Liyanage Chandratilak De Silva, and Pg Emeroylariffion Abas. 2024. Innovating patent retrieval: a comprehensive review of techniques, trends, and challenges in prior art searches. Applied System Innovation, 7(5):91. find
  2. Genrikh Saulovich Altshuller. 1999. The innovation algorithm: TRIZ, systematic innovation and technical creativity. Technical innovation center, Inc. find
  3. Zilong Bai, Ruiji Zhang, Linqing Chen, Qijun Cai, Yuan Zhong, Cong Wang, Yan Fang, Jie Fang, Jing Sun, Weikuan Wang, and 1 others. 2024. Patentgpt: A large language model for intellectual property. arXiv preprint arXiv:2404.18255. — [arXiv:2404.18255](https://arxiv.org/abs/2404.18255) find
  4. Hamid Bekamiri, Daniel S Hain, and Roman Jurowetzki. 2024. Patentsberta: A deep nlp based hybrid model for patent distance and classification using augmented sbert. Technological Forecasting and Social Change, 206:123536. find
  5. Gaetano Cascini and Davide Russo. 2007. Computer-aided analysis of patents and search for triz contradictions. International Journal of Product Development, 4(1-2):52–67. find
  6. Hsiang-Tang Chang, Chen-Yen Chang, and Wen-Kuei Wu. 2017. Computerized innovation inspired by existing patents. In 2017 international conference on applied system innovation (ICASI), pages 1134–1137. IEEE. find
  7. Jianlyu Chen, Junwei Lan, Chaofan Li, Defu Lian, and Zheng Liu. 2025. Reasonembed: Enhanced text embeddings for reasoning-intensive document retrieval. arXiv preprint arXiv:2510.08252. find
  8. Liuqing Chen, Yaxuan Song, Shixian Ding, Lingyun Sun, Peter Childs, and Haoyu Zuo. 2024. Triz-gpt: An llm-augmented method for problem-solving. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, volume 88407, page V006T06A010. American… find
  9. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  10. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  11. Hafsteinn Einarsson, Sigrún Helga Lund, and Anna Helga Jónsdóttir. 2024. Application of chatgpt for automated problem reframing across academic domains. Computers and Education: Artificial Intelligence, 6:100194. find
  12. Stefano Filippi. 2023. Measuring the impact of chatgpt on fostering concept generation in innovative product design. Electronics, 12(16):3535. find
  13. Guillaume Guarino, Ahmed Samet, and Denis Cavallucci. 2022. Patriz: A framework for mining triz contradictions in patents. Expert Systems with Applications, 207:117942. find
  14. Guillaume Guarino, Ahmed Samet, Amir Nafi, and Denis Cavallucci. 2020. Summatriz: summarization networks for mining patent contradiction. In 2020 19th IEEE international conference on machine learning and applications (ICMLA), pages 979–986. IEEE. find
  15. Xingyu Guo, Yi Tan, and Rui Chen. 2025a. Leveraging large language models and triz: A multi-agent framework for automated patent drafting and innovation generation. In International TRIZ and Artificial Intelligence Conference, pages 134–151. Springer. find
  16. Zishun Guo, Meng Song, Xiaofen Fang, Cuiyun Lin, Hengjie Zhang, Xiaoye Li, and Wenxiao Wang. 2025b. Exploring synergies between aigc and triz in the optimisation of road cone design through integrated innovation methods. Journal of Engineering Design, 36(2):256–275. find
  17. Yihan Hou, Manling Yang, Hao Cui, Lei Wang, Jie Xu, and Wei Zeng. 2024. C2ideas: Supporting creative interior color design ideation with a large language model. In Proceedings of the 2024 CHI conference on human factors in computing systems, pages 1–18. find
  18. Imoh M Ilevbare, David Probert, and Robert Phaal. 2013. A review of triz, and its benefits and challenges in practice. Technovation, 33(2-3):30–37. find
  19. Shuo Jiang and Jianxi Luo. 2024. Autotriz: Artificial ideation with triz and large language models. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, volume 88377, page V03BT03A055. American Society of Mechanical Engineers. find
  20. CKM Lee, Jingying Liang, Kai Leung Yung, and Kin Lok Keung. 2024. Generating triz-inspired guidelines for eco-design using generative artificial intelligence. Advanced Engineering Informatics, 62:102846. find
  21. Kevin Ma, Daniele Grandi, Christopher McComb, and Kosa Goucher-Lambert. 2023. Conceptual design generation using large language models. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, volume 87349, page V006T06A021. American Society… find
  22. Achille Souili and Denis Cavallucci. 2012. Toward an automatic extraction of idm concepts from patents. In CIRP Design 2012: Sustainable Product Development, pages 115–124. Springer. find
  23. Achille Souili, Denis Cavallucci, and François Rousselot. 2015. A lexico-syntactic pattern matching method to extract idm-triz knowledge from on-line patent databases. Procedia engineering, 131:418– 425. find
  24. Stefan Trapp and Joachim Warschat. 2024. Llm-based extraction of contradictions from patents. In International TRIZ Future Conference, pages 3–19. Springer. find
  25. Gangfeng Wang, Xitian Tian, Junhao Geng, Richard Evans, and Shengchuang Che. 2016. Extraction of principle knowledge from process patents for manufacturing process innovation. Procedia Cirp, 56:193– 198. find
  26. Shakked Noy and Whitney Zhang. 2023. Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654):187–192. find
  27. Juanyan Wang, Sai Krishna Reddy Mudhiganti, and Manali Sharma. 2024a. Patentformer: a novel method to automate the generation of patent applications. In Proceedings of the 2024 conference on empirical methods in natural language processing: industry track, pages 1361–1380. find
  28. Hyunseok Park, Jason Jihoon Ree, and Kwangsoo Kim. 2013. Identification of promising patents for technology transfers using triz evolution trends. Expert systems with applications, 40(2):736–743. find
  29. Suyuan Wang, Xueqian Yin, Menghao Wang, Ruofeng Guo, and Kai Nan. 2024b. Evopat: A multi-llm-based patents summarization and analysis agent. arXiv preprint arXiv:2412.18100. — [arXiv:2412.18100](https://arxiv.org/abs/2412.18100) find
  30. Orloff Michael. 2006. Inventive thinking through TRIZ: a practical guide. Springer. find
  31. Marissa Radensky, Simra Shahid, Raymond Fok, Pao Siangliulue, Tom Hope, and Daniel S Weld. 2024. Scideator: Human-llm scientific idea generation grounded in research-paper facet recombination. arXiv preprint arXiv:2409.14634. — [arXiv:2409.14634](https://arxiv.org/abs/2409.14634) find
  32. Wenhui Wang, Furu Wei, Li Dong, Hangbo Bao, Nan Yang, and Ming Zhou. 2020. Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transformers. Advances in neural information processing systems, 33:5776–5788. find
  33. Runtao Ren, Jian Ma, and Jianxi Luo. 2025. Large language model for patent concept generation. Advanced Engineering Informatics, 65:103301. find
  34. Shitao Xiao, Zheng Liu, Peitian Zhang, and Niklas Muennighoff. 2023. C-pack: Packaged resources to advance general chinese embedding. Preprint, arXiv:2309.07597.
  35. Eva Sharma, Chen Li, and Lu Wang. 2019. Bigpatent: A large-scale dataset for abstractive and coherent summarization. arXiv preprint arXiv:1906.03741.
  36. Qizhi Xie and Qiang Liu. 2023. Application of triz innovation method to in-pipe robot design. Machines, 11(9):912. find
  37. Homaira Huda Shomee, Zhu Wang, Sathya N Ravi, and Sourav Medya. 2025. A survey on patent analysis: From nlp to multimodal ai. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 8545–8561. find
  38. L Siddharth and Jianxi Luo. 2024. Retrieval augmented generation using engineering design knowledge. Knowledge-Based Systems, 303:112410. find
  39. Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and Tie-Yan Liu. 2020. Mpnet: Masked and permuted pre-training for language understanding. Advances in neural information processing systems, 33:16857– 16867.
  40. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  41. Siyeong Yun, Woojin Cho, Chulhyun Kim, and Sungjoo Lee. 2022. Technological trend mining: identifying new technology opportunities using patent semantic analysis. Information Processing & Management, 59(4):102993. find
  42. Qihao Zhu and Jianxi Luo. 2022. Generative design ideation: a natural language generation approach. In International Conference on-Design Computing and Cognition, pages 39–50. Springer. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-inventive-problem-solving-llms-benchmark/