Using Bug Report Similarity to Enhance Bug Localisation

Bug localisation techniques are proposed as a method to reduce the time developers spend on maintenance, allowing them to quickly find source code relevant to a bug. Some techniques are based on information retrieval methods, treating the source code as a corpus and the bug report as a query. While these have shown success, there remain a number of little-exploited additional sources of information which could enhance the techniques, including the textual similarity between bug reports themselves. Based on successful results in detecting duplicate bug reports, this work asks: if duplicate bugs reports, which by definition are fixed in the same source location, can be detected through the use of similar language, can bugs which are in the same location but not duplicates be detected in the same way? A technique using this information is implemented and evaluated on 372 bugs across 4 projects, and is found to improve performance on all projects. In particular, the technique increases the number of bugs where the first relevant method presented to developers is the first result from 6 to 27, and those in the top-10 from 50 to 57, showing that it can be successfully used to enhance existing bug localisation techniques.