A Novel Approach to Automatic Query Reformulation for IR-based Bug Localization

Automatic query reformulation techniques for Information Retrieval based Bug Localization (IRBL) have been proposed to improve the quality of queries and IRBL performance. Recently proposed techniques determine the quality of queries via the bugs' description and reformulate them using important terms in the top-N source files retrieved by the initial query. However, the bugs' description may not contain enough information about the bugs, and the retrieved top-N files may not always provide important terms. In this paper, we propose a novel automatic query reformulation approach to improve IRBL performance beyond that of a recent technique. Our method expands bug reports using attachments and expands queries by reducing the noisy terms in them. We experimented with 1,546 bug reports. According to our results, we found that the quality of 70 reports was wrongly determined, and our method improved IRBL performance by up to 118% for these reports. Moreover, compared with a state-of-the-art technique, our method resulted in improvements of approximately 17% in Top-1, 11% in MRR@10, and 10% in MAP@10.