Large Language Models for Patent

Team: Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences; Dalian University of Technology

Supervisor: Min Yang, Shiwen Ni, Kan Xu, Yuan Lin

Student Project Leader: Qiyao Wang

Student Member: Hongbo Wang, Huaren Liu

Collections:

News

[2024-12-20] 🥳 The AutoPatent has been fortunate to receive attention and coverage from Xin Zhi Yuan, and it will continue to be expanded and improved in the future.
[2024-12-13] 🎉 We release the first version of AutoPatent.

Introduction

This page showcases our research on applying Large Language Models (LLMs) to the field of intellectual property and patents. Our work includes benchmark development, automatic patent generation, and evaluation frameworks.

Our Work

^*Equal Contribution and ^† Corresponding Author

Preprint

IPBench: Benchmarking the Knowledge of Large Language Models in Intellectual Property

Qiyao Wang, Guhong Chen, Hongbo Wang, Huaren Liu, Minghui Zhu, Zhifei Qin, Linwei Li, Yilin Yue, Shiqiang Wang, Jiayan Li, Yihang Wu, Ziqiang Liu, Longze Chen, Run Luo, Liyang Fan, Jiaming Li, Lei Zhang, Kan Xu, Hamid Alinejad-Rokny, Chengming Li, Shiwen Ni^†, Yuan Lin^†, Min Yang^†

TL;DR: IPBench is a comprehensive benchmark for evaluating LLMs' knowledge in intellectual property domains, covering patent law, trademark, copyright, and trade secrets across multiple languages and jurisdictions.

Website Paper GitHub

Preprint

AutoPatent: A Multi-Agent Framework for Automatic Patent Generation

Qiyao Wang^*, Shiwen Ni^*, Guhong Chen, Huaren Liu, Shule Lu, Xi Feng, Chi Wei, Qiang Qu, Hamid Alinejad-Rokny, Chengming Li, Yuan Lin^†, Min Yang^†

TL;DR: We introduce Draft2Patent, a novel task for generating full-length patents (~17K tokens) from drafts, along with the D2P benchmark. Our AutoPatent framework, leveraging a multi-agent system, excels in patent generation, with Qwen2.5-7B outperforming larger models like GPT-4o and Qwen2.5-72B in metrics and human evaluations.

Website Paper GitHub

Arxiv

IPEval: A Bilingual Intellectual Property Agency Consultation Evaluation Benchmark for Large Language Models

Qiyao Wang, Jianguo Huang, Shule Lu, Yuan Lin^†, Kan Xu, Liang Yang, Hongfei Lin

TL;DR: IPEval introduces a benchmark for assessing Large Language Models' (LLMs) performance in intellectual property (IP) law with 2,657 questions. It evaluates LLMs across key IP areas using zero-shot, 5-few-shot, and Chain of Thought (CoT) approaches. Findings highlight the need for specialized IP LLMs due to language proficiency bias. The benchmark is crucial for developing LLMs with deeper IP knowledge.

Website Paper GitHub HuggingFace