Research On Software Bug Localization Based On Label Propagation

With the increasing of the scale and complexity in software,the possibility of hid-ing bugs in software is becoming larger and larger.The software's abnormal activities will have a serious impact on production and life,so localization and repair of software bugs earlier is very important.Traditional bug localization methods need to analyze the content of bug reports,review source codes,design test cases and execute the tested programs,analyze program variables and execution results,and finally locate buggy s-tatements.Manually accomplishing these tasks will consume a lot of time and energy,and in large-scale software system it is also not feasible.To reduce the burden on de-velopers and reduce labor costs,automated bug localization methods are proposed to locate software bugs with minimal manual intervention,which analyze bug reports and source files,extract features,and locate buggy files through computer technology.Based on the static bug localization,this paper formalizes the bug localization problem as a multi-classification problem to judge whether the source file is related to the bug report and introduces a bug localization framework GBLocator based on label propagation.The framework implements the analysis of bug reports and source files,and then extracts the features between bug reports and bug reports,bug reports and source files,source files and source files.After that,the framework establishs a graph model representing the relationship between bug reports and source files,and locate buggy files by sorting the classification result.The main contributions of this paper are summarized as follows:1.Survey the current research on software bug localization This paper first introduces the process and existing problems of software bug localization.Then it describes the current situation of software bug localization and summarizes the existing software bug localization methods.2.Present a method GBLocator for bug localization based on label propaga-tion.For the automatic bug localization in bug reports,this paper presents a bug local-ization method based on label propagation.The method extracts suitable features from bug reports and source files,then construct a graph which reflects the re-lationship between bug reports and source files and finally optimizes the propa-gation process by minimizing the objective function.The method can effectively use various information to improve the accuracy of software bug localization.3.Perform an empirical study on the effectiveness of GBLocator.In order to verify the effectiveness of the GBLocator method in software bug lo-calization,this paper conducts an empirical study on three open source datasets:swt,aspectj and eclipse,which compares the experimental results with other bug localization methods.In addition,we also design different experiments to study the factors affecting the performance of GBLocator.