Software Defect Prediction Model Based on Attention Mechanism

In recent years, with the gradual expansion of modern software, software components have become more and more complex, and software defects have also increased. The existence of defects often leads to software errors or even crashes, and brings huge economic losses to enterprises. Therefore, how to use software defect prediction technology to find defects as early as possible to improve software reliability has become an important guarantee activity in the process of software projects. Different from the traditional defect prediction methods, this article takes program semantics as the point of penetration, constructs a software defect prediction model based on the attention mechanism on the basis of the abstract syntax tree, and introduces a mask model for the correlation between the function methods in the program files. The evaluation results on the PROMISE data set show that the model proposed in this paper is superior to the two state-of-the-art defect prediction model for both within-project defect prediction and cross-project defect prediction. And it also verifies the necessity of introducing masking model into attention mechanism.