摘要
[目的/意义]为解决当前专利检索模式在检索人员不熟悉专业领域或需优先检索基础专利文献时存在的局限性问题,提出一种基于大模型RAG架构的专利检索Agent及检索模式。[方法/过程]该专利检索Agent支持专利文献库的分级搭建,允许检索人员根据需求和习惯构建个性化的多类型专利文献库,同时采用一种可反馈调整的二次切分向量转化方法提高检索准确度,通过配置Agent的基本功能实现专利检索过程中的互联网数据检索,并借助大模型的生成能力对检索结果进行精炼和输出。[结果/结论]通过两个具体案例,验证该专利检索Agent可实现基础文献库的优先准确检索以及内外网检索的无缝衔接,为提升检索效率以及确保检索结果的准确性提供了全新的视角与策略。
[Purpose/Significance]To address the limitations of current patent retrieval models when retrievers are unfamiliar with the professional field or need to prioritize the retrieval of foundational literature,a patent retrieval Agent and retrieval mode based on the large language model retrieval-augmented generation(RAG)architecture are proposed.[Method/Process]This patent retrieval Agent supports the hierarchical establishment of patent literature databases,allowing retrievers to construct personalized multitype patent literature databases according to their needs and habits.It also adopts a feedback-adjustable secondary segmentation vector transformation method to improve retrieval accuracy.By configuring the basic capabilities of the Agent,Internet knowledge retrieval is achieved,and the retrieval results are refined and output with the help of the large model’s generative capabilities.[Result/Conclusion]Through two specific cases,it is demonstrated that the patent retrieval Agent can achieve accurate and prioritized retrieval of foundational literature databases and seamless integration of LAN and Internet retrieval,providing a new perspective and strategy for improving retrieval efficiency and ensuring the accuracy of retrieval results.
作者
张超然
梁素平
杨杰
杨敏
赵利强
颜斌
ZHANG Chaoran;LIANG Suping;YANG Jie;YANG Min;ZHAO Liqiang;YAN Bin(Patent Examination Cooperation(Beijing)Center of the Patent Office,CNIPA,Beijing 100166;Beijing University of Chemical Technology,Beijing 100029;Baidu.com Times Technology(Beijing)Co.,Ltd.,Beijing 100081)
出处
《中国发明与专利》
2025年第2期14-25,共12页
China Invention & Patent
关键词
大模型
检索增强生成
专利检索
专利文献库
large language model
retrieval-augmented generation(RAG)
patent search
patent literature library
