Research On Software Defect Location Method Based On Code Semantic Features

In the 21 st century,with the rapid development of technological innovation,all walks of life are enjoying the convenience brought by electronic information products.Behind this convenience,many staff members strictly control the quality of each software system,and software defect location is one of the most important steps.Early software defect localization can only rely on the manual experience of developers or testers,and the software defect localization efficiency is low.With the development of Internet technology,people have begun to study how to quickly locate software defects without the developer's experience.Most traditional software defect locating methods use text processing methods to extract vocabulary from software defect reports and source code files,and then use TF-IDF technology and its variants for vectorization.However,most traditional software defect locating methods only consider how often a word appears in documents and corpora.As for the context information between words,the semantics of the code in the software defect report and source code files are not fully considered,and the accuracy needs to be improved.Therefore,this thesis proposes a software defect location method based on code semantic features CSFLoc.Based on the full consideration of software defect reports and source code file code semantics,the convolutional neural network feature extraction is combined with traditional similarity calculation methods to locate software defects.According to the code semantics,the source code file types are divided into entity classes and non-entity classes,and the impact of differentiating source code file types on software defect location is studied.The historical defect report is introduced into the calculation of the similarity between the source code file and the software defect report,and the effect of the historical defect report on software defect location is studied.This article uses the three indicators of Acc@k,MAP,and MRR to evaluate the accuracy of software defect location on four open source projects: SWT,Aspect J,Tomcat,and JDT.The results show that compared with the existing benchmark methods,the Acc@k value,MAP value,and MRR value of the CSFLoc method in this thesis have increased to a certain extent.Studying software defect reports and source code file semantics has improved the software defect location method to a certain extent accuracy.