摘要
[目的/意义]随着人工智能技术的发展,大模型在专利检索领域应用价值凸显。本文旨在提出一种兼具泛化性与精准性的检索范式,以期最大化发挥大模型的作用。[方法/过程]提出“构建提问语义基准—大模型交互—检索要素反哺”三步法检索策略。首先针对不同类型技术方案差异化构建提问语义基准,提出四种构建方法确保大模型初始响应质量;然后与大模型深度交互,引导其输出有价值信息并筛选聚焦检索要素;最后将检索要素反哺至检索平台。[结果/结论]通过四个典型实例,探讨该方法在实践中的应用,验证了方法的有效性,为专利检索工作带来新助力,具有重要实践意义。
[Purpose/Significance]With the rapid advancement of artificial intelligence,the potential of large language models(LLMs)in the field of patent retrieval has become increasingly significant.This paper aims to propose a retrieval paradigm that balances generalizability and accuracy,in order to maximize the potential of LLMs.[Method/Process]A three-step retrieval strategy is proposed:(1)constructing a semantic benchmark for queries,(2)interacting with the LLMs,and(3)feeding back the refined retrieval elements.First,differentiated semantic benchmarks are constructed for various types of technical solutions,and four construction methods are proposed to ensure the initial response quality of the LLMs.Second,we engage in in-depth interaction with the LLMs to guide it to generate valuable information and filter the results to focus on retrieval elements.Finally,the retrieval elements are fed back into the retrieval platform.[Result/Conclusion]Through four representative case studies,the practical application and effectiveness of the proposed method are demonstrated.This method provides a novel impetus for patent retrieval and holds significant practical value.
作者
朱琳玲
于俊
王哲飞
ZHU Linling;YU Jun;WANG Zhefei(Patent Examination Cooperation Jiangsu Center of the Patent Office,CNIPA,Suzhou 215163;Suzhou University of Technology,Suzhou 215000)
出处
《中国发明与专利》
2025年第11期88-98,共11页
China Invention & Patent
关键词
大模型
专利检索
语义基准
交互
反哺
large language models
patent retrieval
semantic benchmark
interaction
feedback
