Treffer: Mining user reviews for method-level bug localization using transformers in java-based applications.

Users frequently post hundreds of reviews on mobile applications, often expressing dissatisfaction, reporting bugs, or suggesting new features. These reviews represent a valuable feedback channel that can be leveraged to improve software quality and user satisfaction. In this study, we propose an enhanced transformer-driven approach to automatically link user reviews to relevant method-level code elements in mobile apps. Rather than stopping at the file or class level, our method identifies specific Java methods that are most likely responsible for the issues described in user feedback. Our pipeline begins by filtering reviews through a combined sentiment and intent-aware layer, where reviews are retained if they either exhibit strong negative sentiment or are classified as complaints or feature requests using a zero-shot intent classification model. To group related reviews, we employ BERTopic. This transformer-based topic modeling technique uses Sentence-BERT embeddings and HDBSCAN clustering to form coherent semantic clusters without predefining the number of topics. Each topic is then represented as a dense vector by averaging the embeddings of the clustered reviews. On the code side, we extract and preprocess Java methods using JavaParser and generate contextual embeddings using Sentence-BERT. Cosine similarity is computed between topic vectors and code vectors to identify the Java methods most closely aligned with the concerns in each cluster. We validate our approach on a dataset of 44,683 user reviews spanning 10 open-source Android applications. The results demonstrate the effectiveness of the method in accurately identifying method-level code elements related to user-reported issues, making it a valuable tool for software maintenance and evolution. [ABSTRACT FROM AUTHOR]

Copyright of Neural Computing & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)