Bug Report Summarization: An Evaluation of Ranking Techniques

Bug reports are regularly consulted software artifacts, especially, because they contain valuable information for many change management tasks. Developers consult them whenever they need to know already reported problems or have to investigate previous bug solutions. This activity, however, consumes a substantial amount of time once bug reports content might achieve dozens of comments and thousands of sentences. One recommended and massively applied solution to prevent developers to read the entire bug report is to summarize the whole conversation in a few sentences. Summaries ideally give to developers an overview of the current status of the bug and the reasons, highlighting the result of each proposed solution, for which environments, which solutions are most appropriated, and the necessary information to reproduce the bug. This strategy intends to minimize the time a developer would spend in maintenance tasks. However, investigations demonstrate that summaries do not meet the expectations of developers and, in practice, they still read the entry line of discussion. To circumvent this drawback, we propose a summary based on comments, instead of the ones based on isolated sentences, as proposed by previous works. We argue that a ranking of the most relevant comments would enable developers to find more appropriate information. Empirical results corroborate with our arguments and show that summaries generated by traditional ranking algorithms are accurate with respect to developers expected information when compared to reference summaries created manually.