期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2021
卷号：99
期号：6
语种：English
出版社：Journal of Theoretical and Applied
摘要：Fixing bug reports is an important activity performed during software maintenance. End-users and software developers report bugs related to open and closed-source projects through a bug tracking system. They should describe the bug reports carefully, mainly when they assign the severity of the bug. Thus, assigning incorrect severity levels will postpone the fixing order of critical bugs. Many works have been proposed using various machine learning algorithms to classify the severity of bug reports. However, few research works have considered the analysis of reporters sentiments expressed in the summary description of bug reports to predict the bug severity. In this paper, the analysis of the reporters sentiments has been considered and incorporated into the severity prediction process. More specifically, sentiment-based approaches have been proposed, namely machine learning and lexicon-based approaches for predicting the severity of bug reports. SentiWordNet dictionary is used to identify the bug reports sentiment terms and compute their associated sentiment scores. The proposed sentiment-based approaches have been applied and evaluated on a bug reports dataset related to closed-source projects extracted from the JIRA bug tracking system. The results of sentiment-based machine learning and lexicon-based approaches are compared and reported. The results have shown that the Logistic Model Trees (LMT) model outperforms other sentiment-based models, including the lexicon-based model. The reported experimental results also revealed that the lexicon-based approach is not effective for bug severity prediction. However, the sentiment-based machine learning approach significantly improves the severity prediction of bug reports compared to the lexicon-based approach (baseline approach). The severity prediction accuracy has been remarkably enhanced from 53% for lexicon-based to 87.14%. Likewise, the F-Measure of the severity prediction has been improved from 0.65 for lexicon-based to 0.91 after applying the machine learning approach.
