Sentiment Analysis for Bug Resolution in Multi-Language Deep Learning Frameworks
Zengyang Li, Tao Jiang, Hui Liu, Sicheng Wang · 2025
In the development of deep learning frameworks (DLFs), bug resolution is an indispensable part of the development process. The bug resolution process often requires numerous developers to coordinate various programming languages and libraries within a complex technical environment to address compatibility and performance issues. This collaborative development environment places significant importance on the sentiment experience of developers throughout the process. Currently, there is a lack of research on the sentiment on bug resolution in multi-language DLFs. To this end, we conducted an empirical study on 10,627 comments related to bugs in three open source multi-language DLFs, namely MXNet, PyTorch, and TensorFlow, to explore the sentiment experience of developers while resolving bugs in multi-language DLFs. The results indicate that (1) developers are generally able to resolve bugs in DLFs with a relatively calm sentiment and typically do not experience extreme negative sentiment. (2) Developers exhibit the lowest positive sentiment when fixing memory bugs and version compatibility bugs, while they show the highest positive sentiment when addressing data bugs and documentation bugs. (3) Developers show higher positive sentiment when fixing single-programming-language bugs compared with multi-programming-language bugs. These findings lay the groundwork for sentiment management in the development of DLFs.