Software Defect Prediction for Specific Defect Types based on Augmented Code Graph Representation

In a software life cycle, improving quality and identifying and repairing defects has become an important research topic. Previous studies have proposed defect prediction based on artificial measurement features, a method whose quality is unfortunately difficult to guarantee. On the other hand, many current studies have attempted to predict all types of defects using a single model, which is difficult to achieve. In this paper, Augmented-CPG, a new code graph representation, is proposed. Based on this representation, a defect region candidate extraction method related to the defect type is proposed. Graphic neural networks are introduced to learn defect features. We carried out experiments on three different types of defects, and the results show that our method can effectively predict specific types of defects.