Is this bug severe? A text-cum-graph based model for bug severity prediction

TL;DR

This paper presents a novel text-cum-graph neural model for predicting bug severity in large software repositories, combining bug descriptions, comments, and bug relation graphs for improved accuracy.

Contribution

It introduces a combined text and graph neural network approach for bug severity prediction, leveraging bug descriptions, comments, and bug relation graphs.

Findings

Effective bug severity prediction using combined text and graph models

Improved accuracy over text-only models

Demonstrated benefits of incorporating bug relation graphs

Abstract

Repositories of large software systems have become commonplace. This massive expansion has resulted in the emergence of various problems in these software platforms including identification of (i) bug-prone packages, (ii) critical bugs, and (iii) severity of bugs. One of the important goals would be to mine these bugs and recommend them to the developers to resolve them. The first step to this is that one has to accurately detect the extent of severity of the bugs. In this paper, we take up this task of predicting the severity of bugs in the near future. Contextualized neural models built on the text description of a bug and the user comments about the bug help to achieve reasonably good performance. Further information on how the bugs are related to each other in terms of the ways they affect packages can be summarised in the form of a graph and used along with the text to get additional benefits.