Issue reports are valuable resources for the continuous maintenance and improvement of software. Managing issue reports requires a significant effort from developers. To address this problem, many researchers have proposed automated techniques for classifying issue reports. However, those techniques fall short of yielding reasonable classification accuracy. We notice that those techniques rely on text-based unimodal models. In this paper, we propose a novel multimodal model-based classification technique to use heterogeneous information in issue reports for issue classification. The proposed technique combines information from text, images, and code of issue reports. To evaluate the proposed technique, we conduct experiments with four different projects. The experiments compare the performance of the proposed technique with text-based unimodal models. Our experimental results show that the proposed technique achieves a 5.07% to 14.12% higher F1-score than the text-based unimodal models. Our findings demonstrate that utilizing heterogeneous data of issue reports helps improve the performance of issue classification.

Issues are evolving requirements in software engineering. They are the main factors that increase the cost of software evolution. To help developers manage issues, GitHub provides issue labeling mechanisms in issue management systems. However, manually labeling issue reports still requires considerable developer workload. To ease developers’ burden, researchers have proposed automatically classifying issue reports. To improve the classification accuracy, researchers adopted deep learning techniques and pretrained models. However, pretrained models in the general domain such as RoBERTa have limitations in understanding the contexts of software engineering tasks. In this paper, we create a pretrained model, IssueBERT, with issue data to understand whether a domain-specific pretrained model could improve the accuracy of issue report classification. We also adopt and explore several pretrained models in the software engineering domain, namely, CodeBERT, BERTOverflow, and seBERT. We conduct a comparative experiment on these pretrained models to evaluate their performance in classifying issue reports. Our comparison results show that IssueBERT outperforms the other pretrained models. Noticeably, IssueBERT yields an average F1 score that is 1.74% higher than that of seBERT and 3.61% higher than that of RoBERTa, even though IssueBERT was pretrained with much less data than seBERT and RoBERTa.

Software continuously evolves through changes, and issue reports encapsulate these change requests. In the GitHub system, a labeling mechanism has been introduced for systematic issue management, but significant effort from developers is required to label and manage these issues. To address this, numerous attempts have been made in previous research to automate issue report classification. However, these attempts have shown limitations in classification accuracy. We experiment to determine if integrating heterogeneous information through a multimodal model that combines text, images, and code from issue reports can improve classification accuracy. Specifically, we investigate whether training the model on extensive issue data can enhance classification accuracy. Experimental results show that the multimodal approach outperforms single-modal models by 5.50-7.01% in terms of F1-Score, demonstrating superior performance. These findings indicate that leveraging heterogeneous data sources in issue reports is effective in improving classification performance.