A Hierarchical Attention Networks based Model for Bug Report Prioritization
Anurag Yadav, Santosh Singh Rathore · 2024
Software is becoming increasingly complex, leading to a surge in bugs. Bug reports typically contain essential details for bug reproduction and resolution. Currently, a triage engineer manually reads bug reports, classifies the bugs, and assigns priorities to the critical bugs. This task is time-consuming. While machine learning and deep learning models have been developed to automate this task, some have shown subpar performance, and others, although promising, have demonstrated inconsistency when applied to new bug reports. This paper proposes a Hierarchical Attention Network (HAN) model for prioritizing software bug reports. The bug reports are preprocessed through text normalization, tokenization using the DistilBERT tokenizer, and attention mechanisms to capture hierarchical relationships within the text. The model architecture incorporates bidirectional GRU layers and attention mechanisms at both word and sentence levels. The proposed model is evaluated on four software projects using precision, recall, and F1-score measures. Next, the proposed model is compared with convolution neural networks (CNN), support vector machines (SVM), random forests, and logistic regression techniques. The results showed that the proposed HAN model performed better than the other techniques.