An Automated Approach for the Prediction of the Severity Level of Bug Reports Using GPT-2

Abstract

Manual investigation is warranted in traditional approaches for estimating the bug severity level, which adds to the effort and time required. For bug severity report prediction, numerous automated strategies have been proposed in addition to manual ones. However, the current bug report predictors by facing several issues, such as overfitting and weight computation, and therefore, their efficiency for specific levels of data noise needs to improve. As a result, a bug report predictor is required to solve these concerns (e.g., overfitting and avoiding weight calculation, which increases computing complexity) and perform better in the situation of data noise. We use GPT-2’s features (limiting overfitting and supplying sequential predictors rather than weight computation) to develop a new approach for predicting the severity level of bug reports in this study. The proposed approach is divided into four stages. First, the bug reports are subjected to text preprocessing. Second, we assess each bug report’s emotional score. Third, each report is presented in vector format. Finally, an emotion score is assigned to each bug report, and a vector of each bug report is produced and sent to GPT-2. We employ statistical indicators like recall, precision, and F1-score to evaluate the suggested method’s effectiveness and efficacy. A comparison was also made using state-of-the-art bug report predictors such as Random Forest (RF), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) Network, Support Vector Machine (SVM), XGBoost, and Naive Bayes Multinomial (NBM). The proposed method’s promising result indicates its efficacy in bug information retrieval.