Learning Graph Representation of Bug Reports to Triage Bugs using Graph Convolution Network

Abstract

Bug triage is a software engineering problem, which is being solved by classification methods. The social network analysis, mining repositories, statistical modeling, topic modeling, machine learning, and deep learning techniques have been used to triage the bugs. These existing methods showed promising results but still far from perfection, which requires improvement. This paper proposes a graph representation method for the bug reports dataset, which solves the bug triage problem with the node classification problem. The heterogeneous graph is built using the word to word and word to document co-occurrences for the whole bug dataset. The graph convolution network (GCN) is trained on the generated graph to learn the bug reports' graph representation. The proposed method is validated on the open-source project's bug data. Top-K accuracy is used as an evaluation metric to evaluate the performance of the model. The reported results show promising results compared to previous studies.