Using BERT to Predict Bug-Fixing Time
Pasquale Ardimento, Costantino Mele
Abstract
Prediction of the resolution time of a newly-submitted bug is a relevant aspect during the bug triage process since it can help project managers to better estimate software maintenance efforts and better manage software projects. Once a bug is reported, it is typically recorded in a Bug Tracking System, and it is assigned to a developer in order to solve the issue. The contribution of this paper is to provide a deep learning approach for the resolution of the bug-fixing time prediction, proposing a new feature set, consisting of the description of the issue and comments of the developers, in order to perform transfer learning on a pre-trained language representations model, called BERT. The problem of predicting the resolution time of a bug is therefore formulated as a supervised text categorization task. BERT makes use of a self-attention mechanism that allows learning the bidirectional context representation of a word in a sentence, which constitutes one of the main advantages over the previously proposed solutions. Experimental results show the proposed approach has effective bug-fixing time prediction ability.