Abstract
Software bug localization is very important in software engineering, but it is also complicated and time consuming. To improve the efficiency of developers, researchers have developed various traditional bug localization and machine learning bug localization methods. In this paper, we propose a novel method that improves bug localization performance. First, surface lexical correlation matching between bug reports and source code files is used to obtain features by deep neural network. Second, to solve the lexical gap between bug reports and source code files, semantic correlation matching between them is used to obtain features based on word embedding and sentence embedding by deep neural network. Then, the joint features obtained by the surface lexical and semantic correlation matching are fused into a unified feature representation for bug reports and source code files. In addition, since our experimental datasets are imbalanced data, we use a focal loss function to solve the impact of data imbalance. Finally, our method obtains the relatively high bug localization performance compared to other classic methods.
Highlights
Software defect fixes in the software lifecycle have always been very important
Xiao et al [13] presented the DeepLoc method, which obtains vector of bug reports and source code files by word2vec and sent2vec to locate relevant buggy files by convolutional neural networks. In these bug localization methods based on deep learning, some consider surface lexical correlation matching between bug reports and source code files, regardless of their semantic correlation matching
The main contributions of our work are as follows: 1. We propose a novel method called joint surface lexical and semantic correlation matching based on convolutional neural networks (SLS-CNN) for bug localization
Summary
Software defect fixes in the software lifecycle have always been very important. General software defects are fed back to the development team by bug reports, and the development team fixes the bugs based on the reports. Wang and Lo [7] proposed the AmaLgam+ method, which uses similar report, structure, and other information to locate relevant buggy files These traditional methods use feature attributes of bug reports and source code files for bug localization. Xiao et al [13] presented the DeepLoc method, which obtains vector of bug reports and source code files by word2vec and sent2vec to locate relevant buggy files by convolutional neural networks. In these bug localization methods based on deep learning, some consider surface lexical correlation matching between bug reports and source code files, regardless of their semantic correlation matching.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
