A Method of Component Prediction for Crash Bug Reports Using Component-Based Features and Machine Learning

In large-scale software with hundreds of components, assigning a bug report to a relevant component representing a group of developers who can analyze and solve the potential issue can be a challenging task. In such a practical environment, we observed that bug reports often get re-assigned multiple times because of incorrect root cause component identification. Each re-assignment considerably delays the bug fix time. Therefore, it is desirable to identify a suitable component correctly. We report on our experience with component prediction when call stacks are used as the primary data of bug reports. Our method extracts the component features by aggregating the frequency of component functions from the call stack after introducing the component name and position into the original call stack. By predicting the component position of the call stack using component features and machine learning, we can obtain a suitable component of the bug report from the call stack. Our method achieves 73% accuracy on crash bug reports for SAP HANA, a large industrial in-memory database system. It significantly reduces the time delay for assigning bug reports to suitable components.