Cost-Aware Clustering of Bug Reports by Using a Genetic Algorithm

摘要

The inefficient distribution of bugs to developers is increasing the cost of software development and maintenance. In efforts to tackle this issue, various studies have been carried out to recommend suitable developers for specific bugs. These studies often leverage similarity between bug reports; for example, if a developer addressed a bug report similar to a newly incoming report, that developer can be suitable to fix the bug described in the new report. However, the existing studies have resulted in imbalanced distribution - a large number of bugs can be concentrated in a small number of developers. In this paper, we propose a novel approach to achieve a cost-aware distribution of bug reports to support workload balancing. Our approach is composed of two phases. First, a set of similar report groups composed of strongly related bugs is generated based on their similarity and dependency. Clusters are then created by grouping the similar report groups so that each cluster can have similar cost (i.e., minimizing its standard deviation). Our approach leverages a genetic algorithm to find a near-optimal distribution of bug reports because it is an NP-hard problem. The experiments with 1,047 bug reports collected from Mozilla's Firefox were conducted to evaluate our approach. The results showed that our approach effectively provides an appropriate solution to achieve a cost-balanced distribution of bug reports. In addition, we carried out a user study targeting 30 developers from 15 companies to figure out the usefulness and effectiveness of our approach. Among the participants, 67% answered that our approach is useful for triaging their bugs to developers. This shows the possibility for use in cases of managing or triaging bugs from the project manager's perspective.