知识图谱和大语言模型在航天测控问答系统中的融合应用研究

Fusion Application of Knowledge Graph and Big Language Model to Aerospace TT&C Question-answering System

随着自然语言处理技术的发展,智能化检索与问答系统不断发展,为改善传统知识图谱语义解析能力不足和当前通用大语言模型在垂直领域对知识学习不够深入的问题,提出了融合知识图谱的大语言模型方法,进行了两步优化:首先,利用知识图谱在命名实体识别和关系抽取的基础上,构建大模型提示Prompt模板,进行辅助增强生成,利用图谱中存储的数据提供相关来源;其次,利用低阶适应性微调(LoRA)策略冻结大模型原有参数,增加部分网络参数进行微调训练,优化模型在航天测控领域的知识储备与理解。通过两步改进提高了模型整体在语义解析和知识细节上的理解掌握,结合航天测控领域的相关教材、报告和手册等资料,搭建了知识问答系统,取得了较好的效果,说明了该方法具有一定的应用价值。

With the development of natural language processing technology,the intelligent retrieval and questionanswering system continuously developed.In order to improve the semantic parsing ability of traditional knowledge graph and the problem that the current general large language model is not deep enough for knowledge learning in the vertical domain,a large language model method integrating knowledge graph is proposed,and a two-step optimization is carried out.First,on the basis of named entity recognition and relationship extraction,Build a large model Prompt template for auxiliary enhancement generation,using the data stored in the map to provide relevant sources.Second,the low-rank adaptation(LoRA)strategy is used to freeze the original parameters of the large model,and some network parameters are added for fine-tuning training,so as to optimize the knowledge reserve and understanding of the model in the field of aerospace tracking,telemetering,and command(TT&C).Through the two-step improvement,the overall semantic analysis and knowledge details of the model are improved.Combined with the relevant textbooks,reports,and manuals in the field of aerospace TT&C,a knowledge question-answering system is built,and good results are obtained,indicating that the method has certain application value.