Automatic Classification of Bug Reports Based on Multiple Text Information and Reports’ Intention

Abstract

With the rapid growth of software scale and complexity, a large number of bug reports are submitted to the bug tracking system. In order to speed up defect repair, these reports need to be accurately classified so that they can be sent to the appropriate developers. However, the existing classification methods only use the text information of the bug report, which leads to their low performance. To solve the above problems, this paper proposes a new automatic classification method of bug reports. The innovation is that when categorizing bug reports, in addition to using the text information of the report, the intention of the report (i.e. “suggestion” or “explanation”) is also considered, thereby improving the performance of the classification. First, we collect bug reports from four ecosystems (Apache, Eclipse, Gentoo, Mozilla) and manually annotate them to construct an experimental data set. Then, we use Natural Language Processing technology to preprocess the data. On this basis, BERT and TF-IDF are used to extract the features of the intention and the multiple text information. Finally, the features are used to train the classifiers. The experimental result on five classifiers (including K-Nearest Neighbor, Naive Bayes, Logistic Regression, Support Vector Machine and Random Forest) show that our proposed method achieves better performance and its F-Measure achieves from 87.3% to 95.5%.