Improving the Robustness of the Bug Triage Model through Adversarial Training
Min-Ha Kim, Dae-Sung Wang, Sheng-Tsai Wang, Seo-Hyeon Park, Chan-Gun Lee · 2022 International Conference on Information Networking (ICOIN) · 2022
Recently, research on automated bug triage using deep neural networks is being actively conducted. Unfortunately, there have been few studies on improving the robustness of the bug triage models through adversarial training. In this paper, we present an approach to evaluate and improve the robustness of the bug triage model. We present a new method for generating adversarial bug reports. We exploit the test coverage to compare the robustness of various models for bug triage. The experimental results suggest that our model has better robustness. In addition, the proposed technique for adversarial data generation is superior compared to the existing techniques in three aspects: adversarial data generation time, document similarity, and word change rate.