Automatic Knowledge Graph Feature Construction for Software Requirement Documents

Software requirements documentation is usually written in descriptive text and it is conducive to converting unstructured requirements documents into knowledge graphs for requirement management and reuse. However, there are certain challenges in the automatic construction of knowledge graphs for software requirements documents because there are various generalized entities and entity relationships in the documents, and entity relationship extraction cannot be performed through existing knowledge bases or predefined relationship types. This paper proposes a method to construct knowledge graph for software requirements documents automatically. Based on the characteristics of the requirement text, a new sequence labeling scheme and a BERT-BIGRU sequence labeling model are proposed. The BERT-BIGRU model is used to recognize requirement entities and extract entity relationship jointly, which can be used to construct a knowledge graph. The experimental results show that the proposed method can effectively realize the automatic construction of software requirement knowledge graph.