Using word embedding and convolution neural network for bug triaging by considering design flaws

Resolving bugs in the maintenance phase of software is a complicated task. Bug assignment is one of the main tasks for resolving bugs. Some Bugs cannot be fixed properly without making design decisions and have to be assigned to designers to avoid emerging bad smells that may cause subsequent bug reports. Hence, it is important to refer some bugs to the designer to check the possible design flaws. Based on our best knowledge, no work has considered referring bugs to designers. This issue is considered in this work. In this paper, a dataset is created, and a CNN-based model is proposed to predict the need for assigning a bug to a designer by learning the peculiarities of bug reports effective in creating bad smells in the code. The features of each bug are extracted from CNN based on its textual features, such as a summary and description. The number of bad samples added to it in the fixing process using the PMD tool determines the bug tag. The summary and description of the new bug are given to the model and the model predicts the need to refer to the designer to review the design. An accuracy of 75% (or more) was achieved for datasets with a sufficient number of samples for deep learning-based model training. A model is proposed to predict bug referrals to the designer. The efficiency of the model in predicting referrals to the designer at the time of receiving the bug report was demonstrated by testing the model on 10 projects.