NL-based query refinement and contextualized code search results: A user study

As software systems continue to grow and evolve, locating code for software maintenance tasks becomes increasingly difficult. Source code search tools match a developer’s keyword-style or natural language query with comments and identifiers in the source code to identify relevant methods that may need to be changed or understood to complete the maintenance task. In this search process, the developer faces a number of challenges: (1) formulating a query, (2) determining if the results are relevant, and (3) if the results are not relevant, reformulating the query. In this paper, we present a NL-based results view for searching source code for maintenance that helps address these challenges by integrating multiple feedback mechanisms into the search results view: prevalence of the query words in the result set, results grouped by NL-based information, as a result list, and suggested alternative query words. Our search technique is implemented as an Eclipse plug-in, CONQUER, and has been empirically validated by 18 Java developers. Our results show that users prefer CONQUER over a state of the art search technique, requesting customization of the interface in future query reformulation techniques.