Multi-scale Convolutional Neural Networks-Based Severity Prediction

The software defect report is an important part of the software repair process, and its severity attribute is very important. According to the description of the severity attribute of the software defect report, it can assist the developer to solve the defects described in the important defect report within a limited time. But manually assessing the severity is a tedious task. In this paper, we use a multi-scale convolutional neural network algorithm to automatically predict the severity attribute of defect reports. First, we use natural language processing technology to preprocess the defect report. Then create a vector for the defect report. Finally, the constructed vector input classifier based on multi-scale convolutional neural network is used to predict the severity. By comparing with other severity prediction algorithms, it can be found that our method improves the accuracy of the severity prediction of defect reports.