Software developers use the information from software bug reports to correct software defects and enhance software. However, it is difficult for software developers to manually inspect and analyze a “large number of bug reports.” Researchers have thus developed ways to automatically analyze bug reports, including automated bug report classification.

The authors of this paper propose a feature set for automated bug report classification: “the occurrences of certain keywords in the bug reports” are used as features “to classify bug reports into corrective and perfective classes.” This approach is then used to classify “bug reports from three open source projects” with three different classifiers: naive Bayes (NB), support vector machines (SVMs), and random trees (RTs). Their experiments show that a SVM with the proposed feature set produced the highest average classification accuracy.

The authors’ contribution is a new feature set for classifying bug reports, which is more efficient than using the whole document-term matrix. The authors could provide more details on the manual labeling process and the datasets. More performance metrics--recall, precision, and the F1 score--should be calculated to compare the performances of the different classifiers. The authors could also compare automated bug report classification performance with both the proposed feature set and the document-term matrix.