| 摘要: |
| 当前电子信息领域积累的越来越多宝贵经验知识对知识使用技术提出了新的挑战。知识图谱（Knowledge Graph，KG）技术和大规模预训练语言模型(Large Language Model,LLM)技术在知识使用方面都各自存在缺陷，但两种技术的优缺点能够形成互补。因此，基于LLM技术，提出了应用于电子信息领域的知识图谱自动构建与检索增强问答技术。首先基于LLM的语义理解能力自动构建电子信息领域知识图谱，然后构建基于知识图谱和检索增强大模型的知识问答系统。在CoNLL2003数据集和构建的电子信息领域数据集上的实验证明了所方法具有较好质量，知识问答系统具有较好的实用效果。所提方法能够更好地满足从业人员从海量文档中提取相关知识，提高知识利用效率的迫切需求，为推动大模型结合知识图谱技术在电子信息垂直领域的落地应用提供参考。 |
| 关键词: 电子信息领域 知识图谱构建 检索增强 大模型 |
| DOI：10.20079/j.issn.1001-893x.240422003 |
|
| 基金项目: |
|
| Automatic Construction and Retrieval of Knowledge Graph in Electronic Information Field Based on Large Language Model(LLM) |
| XIE Minghua |
| (Southwest China Institute of Electronic Technology，Chengdu 610036，China) |
| Abstract: |
| The increasing accumulation of valuable experiential knowledge in the field of electronic information poses new challenges for knowledge utilization technologies.Both of Knowledge Graph (KG) technolog and Large Language Model(LLM) have their respective shortcomings in knowledge utilization.In view of the fact that the strengths and weaknesses of LLMs and KGs can complement each other,the author proposes a knowledge graph automatic construction and retrieval-enhanced question-answering technology based on LLMs for the electronic information field.This approach first automatically constructs a knowledge graph for the electronic information field based on the semantic understanding capabilities of LLMs and then builds a knowledge question-answering system based on the knowledge graph and retrieval-enhanced large models.Experiments on the CoNLL2003 dataset and a dataset constructed for the electronic information field demonstrate that the proposed knowledge graph automatic construction method is of good quality and that the knowledge question-answering system is effective in practical applications.The proposed method better meets the urgent demand for professionals to extract relevant knowledge from massive documents and improve knowledge utilization efficiency without requiring extensive annotation work.It is hoped that this method will provide a reference for promoting the integration of large models and knowledge graph technologies in vertical applications in the electronic information field. |
| Key words: electronic information knowledge graph construction large language model retrieval augmented large language model(LLM) |
