On Fusing Artificial and Convolutional Neural Network Features for Automatic Bug Assignments

Abstract

Automated bug report assignment is critical for large-scale software projects where reported bugs are frequent and expert developers are required to fix them on time. Finding an appropriate developer with the necessary skill sets and prior experience in fixing similar bugs is difficult and can be an expensive process, depending on the severity of the reported bug. To address this issue, researchers have proposed several machine learning and deep learning-based automated bug report assignment techniques that make use of historical data on reported bugs as well as fixer information. However, there is still room for improvement in the performance of these techniques. In this paper, we propose a novel deep learning-based approach that utilizes two sets of features from the reported bugs’ textual data, namely contextual information and the occurrence of repeating keywords. We develop convolutional neural network and artificial neural network modules to mine these features. We then fuse these two sets of extracted features to assign a bug to an appropriate developer. We conduct extensive experiments on eight benchmark datasets of open-source, real-world software projects to assess the effectiveness of our approach. The experimental results demonstrate that our information fusion-based approach outperforms previous models and improves automated bug report assignment. Furthermore, we debug the errors of our proposed model and publish all source code so that future researchers can contribute to this problem.

Keywords

• Artificial neural network
• bug report assignment
• convolutional neural network
• deep learning
• dimensionality reduction