An approach to generate the bug report summaries using two-level feature extraction

Abstract

Bug report is one of the major software artifact which is generated during the software development process. Changing requirements in the software development process leads to the continuous evolution of bugs which give challenges to the project management task. Bug Reports are the most consulted artifact by the software community. A Bug Report not only contains the information about the bug but also includes information like the resolution process, the enhancements by other persons, and the suggestions from the users if there are any. During the software evolution and maintenance phase, a developer spends a lot of effort and time searching for the appropriate bug report for resolving the bug quickly. Automatic Bug Report Summarization is one approach to solve the issue of time and effort. Bug report summarization helps developers not only find the appropriate bug report quickly but also assists in managing many tasks related to Bug Report Maintenance. In this paper, we have developed a two-level approach to generate the Bug Report summaries where the title, the description and the comments of a resolved bug report are considered for the summary. We find the entities in the title to create a template-based sentence for describing what the bug report is about. We use the PageRank algorithm along with the cosine similarity measure to find the summary of the Description field of bug report. The two level feature-based approach is used to find the relevant comments and the sentences from the comments. We have used the BRC dataset which has been used by most of the research community in this field. Finally the summaries from title, description and comments are merged together to create a final summary. The approach uses the features which have been successfully used by other researchers for the text summarization and especially to the domain of the meeting conversation like data. Empirical results shows that our approach works equally well with the other supervised and unsupervised approaches in terms of ROUGE Scores.