| Summary: | Bug reporting and fixing the reported bugs play a critical part in the development and maintenance of software systems. The software developers and end users can collaborate in this process to improve the reliability of software systems. Various end users report the defects they have found in the software and how these bugs affect them. However, the same defect may be reported independently by several users leading to a significant number of duplicate bug reports. There are a number of existing methods for detecting duplicate bug reports, but the best results so far account for only 24% of actual duplicates. In this paper, we propose a new method based on clustering to identify a larger proportion of duplicate bug reports while keeping the false positives of misidentified non-duplicates low. The proposed approach is experimentally evaluated on a large sample of bug reports from three public domain data sets. The results show that this approach achieves better performance in terms of a harmonic measure that combines true positive and true negative rates when compared to the existing methods.
|
|---|