基于需求文档和图神经网络的需求知识图谱构建方法

Construction Method of Requirement Knowledge Graph Based on Requirement Document and Graph Neural Network

摘要: 知识图谱可以将大量不同种类信息连成一个语义网络，为人工智能相关的研究提供了从“关系”层面解释分析问题的能力。因此，构建各领域知识图谱成了近年来的研究热点。然而对于服务需求领域知识图谱的构建研究较少，针对这一现状，本文提出了一种基于需求文档和图神经网络的需求知识图谱构建方法，该方法以100条公司真实需求文档为数据源，结合传统自然语言处理方法和图神经网络分类模型，通过知识抽取、人工标注、需求特征编码、需求分类、需求图谱存储和可视化等5个步骤来进行服务需求领域知识图谱的构建。实验表明该方法可以有效地从大量非结构化需求文档中提取到需求语义信息，并通过图神经网络分类模型较准确地识别需求意图，从而结合图数据库和可缩放矢量图可视化技术将需求图谱进行轻量级存储和可视化展示。

Abstract: The knowledge graph can connect a large number of different kinds of information into a semantic network, and provides the ability to explain and analyze problems from the “relationship” level for the AI related research. Therefore, the construction of knowledge graph in various fields has become a research hotspot in recent years. But there are few ways to build a requirement knowledge graph. In order to solve this problem, this paper proposes a requirement knowledge graph construction method based on requirement document and graph neural network. The method takes 100 real company requirement documents as the data source. Combined with the traditional natural language processing method and the graph neural network classification model, through knowledge extraction, manual annotation, requirement feature coding, requirement classification, requirement graph storage and visualization, the knowledge graph for the requirements domain will have been successfully constructed. Experiments show that this method can effectively extract requirement semantic information from a large number of unstructured requirement documents, and identify requirement intention more accurately through the graphical neural network classification model. Thus, combined with graph database and scalable vector graph of visualization technology, the requirement graph is lightweight-stored and visualized.