Laprob: A Label propagation-Based software bug localization method

摘要

Context: Bug localization, which locates suspicious snippets related to the bugs mentioned in the bug reports, is time-consuming and laborious. Many automatic bug localization methods have been proposed to speed up the process of bug fixing and reduce the burden on developers. However, these methods have not fully utilized the intra-relations and inter-relations among the bug reports and the source files (i.e., call relationships between the source files).Objective: In this paper, we propose a novel method LaProb (a label propagation-based software bug localization method) that makes full use of the intra-relations and inter-relations among the bug reports and the source files.Method: LaProb transforms the problem of bug localization into a multi-label distribution learning problem. LaProb first constructs a BHG (Biparty Hybrid Graph) by analyzing the structures and contents of bug reports and source files, and calculates the intra-relations between pairs of bug reports and source files, as well as the interrelations between bug reports and source files. Based on BHG, LaProb then predicts the label distribution on source files by using the label propagation algorithm for the target bug report. Finally, LaProb finishes the bug localization task by sorting the results of label propagation.Results: The experimental results on nine open-source software projects (i.e., SWT, AspectJ, Eclipse, ZXing, SEC, HIVE, HBASE, WFLY and ROO) show that compared with several state-of-the-art methods (including BugLocator, BRTracer, BLUiR, AmaLgam, Locus and BLIZZARD), LaProb performs the best in terms of all five metrics on average. For MAP performance measure, LaProb achieves an improvement of 30.9%, 36.6%, 28.0%, 22.2%, 20.1% and 53.5%, respectively.Conclusion: LaProb is capable of making full use of the intra-relations and inter-relations among the bug reports and the source files and achieves better performance than seven state-of-the-art methods.