Sentiment-Based Neural Network Approach for Predicting the Severity of Bug Reports
During the software maintenance process, bugs encountered by software users need to be solved according to their severity level to improve the quality of the software. Therefore, bug reports with high severity should have the highest priority to be fixed. A numerous number of bug reports are submitted daily through Bug Tracking Systems (BTS) such as Bugzilla. The bug triager examines the incoming bug reports manually and verifies whether the assigned severity level is correct or not. This manual process is time-consuming, requires much effort and is possibly error-prone. However, a limited number of research works have considered the sentiments of the bug reporters in predicting the severity of bug reports. This paper proposes two approaches based on sentiment analysis, the Lexicon-based and Multilayer Perceptron (MLP) neural network approaches. The sentiment analysis process determines and measures the sentiment words and their sentiment scores according to the popular sentiment lexicon called SentiWordNet. The proposed approaches are validated on the Eclipse open-source project, and the sentiment-based models performance is evaluated. According to the experimental results, the sentiment-based MLP outperforms the Lexicon-based approach (baseline approach). The F-measure has been improved significantly from 0.52 for the Lexicon-based approach to 0.81 when applying the MLP approach.
5th International Conference on Intelligent Computing in Data Sciences, ICDS 2021
Baarah, Aladdin; Aloqaily, Ahmad; Zyod, Hala; and Mustafa, Nasser, "Sentiment-Based Neural Network Approach for Predicting the Severity of Bug Reports" (2021). Kean Publications. 1043.