Scholarlink

Against the backdrop of the “dual carbon” goals, in order to fully utilize the patent knowledge in the field of battery energy storage and support technological innovation and industrial decision-making, a framework for constructing and applying the battery energy storage patent knowledge graph that integrates machine learning and large language models is proposed: The data layer is constructed with China Patent Network as the basic data source; the ontology framework is built by integrating domain experts’ knowledge and relevant research; after conducting multi-model comparison experiments, the PURE model is used to extract five types of valuable entities (device, material, system, energy, and technology) and their associated relationships from unstructured patent texts in the field of battery energy storage. Knowledge storage and visualization are realized through Neo4j, and an intelligent question-answering system is developed by integrating the DeepSeek-R1 large language model to achieve accurate querying of entities and associated explanations. The research results show that the F1 values of PURE in entity and its relationship extraction tasks reach 90.78 and 68.60% respectively. A knowledge graph containing 9175 entity nodes and 1770 relationship edges has been successfully constructed, and a question-answering system based on the large language model has been developed. The research results can provide support and reference for the analysis of technological innovation paths and the layout of patent strategies in the field of battery energy storage.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Construction and Application of Patent Knowledge Graph in Battery Energy Storage Field by Integrating Machine Learning and Large Language Models

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年

查看全文
  • Xu Hao,
  • Hua Chongji,
  • Ye Sheng,
  • Long Ying,
  • Kang Zhenyuan,
  • Shi Junkan,
  • Zhu Kang

摘要

Against the backdrop of the “dual carbon” goals, in order to fully utilize the patent knowledge in the field of battery energy storage and support technological innovation and industrial decision-making, a framework for constructing and applying the battery energy storage patent knowledge graph that integrates machine learning and large language models is proposed: The data layer is constructed with China Patent Network as the basic data source; the ontology framework is built by integrating domain experts’ knowledge and relevant research; after conducting multi-model comparison experiments, the PURE model is used to extract five types of valuable entities (device, material, system, energy, and technology) and their associated relationships from unstructured patent texts in the field of battery energy storage. Knowledge storage and visualization are realized through Neo4j, and an intelligent question-answering system is developed by integrating the DeepSeek-R1 large language model to achieve accurate querying of entities and associated explanations. The research results show that the F1 values of PURE in entity and its relationship extraction tasks reach 90.78 and 68.60% respectively. A knowledge graph containing 9175 entity nodes and 1770 relationship edges has been successfully constructed, and a question-answering system based on the large language model has been developed. The research results can provide support and reference for the analysis of technological innovation paths and the layout of patent strategies in the field of battery energy storage.