Predicting Bug-Fix Time: Using Standard Versus Topic-Based Text Categorization Techniques

In modern software development, finding and fixing bugs is a vital part of software development and quality assurance. Once a bug is reported, it is typically recorded in the Bug Tracking System, and is assigned to a developer to resolve (bug triage). Current practice of bug triage is largely a manual collaborative process, which is often time-consuming and error-prone. Predicting on the basis of past data the time to fix a newly-reported bug has been shown to be an important target to support the whole triage process. Many researchers have, therefore, proposed methods for automated bug-fix time prediction, largely based on statistical prediction models exploiting the attributes of bug reports. However, existing algorithms often fail to validate on multiple large projects widely-used in bug studies, mostly as a consequence of inappropriate attribute selection [2]. In this paper, instead of focusing on attribute subset selection, we explore an alternative promising approach consisting of using all available textual information. The problem of bug-fix time estimation is then mapped to a text categorization problem. We consider a multi-topic Supervised Latent Dirichlet Allocation (SLDA) model, which adds to Latent Dirichlet Allocation a response variable consisting of an unordered binary target variable, denoting time to resolution discretized into FAST (negative class) and SLOW (positive class) labels. We have evaluated SLDA on four large-scale open source projects. We show that the proposed model greatly improves recall, when compared to standard single topic algorithms.