SNA-DeepRec: Integrating Developer Social Networks and Deep Representation Learning for Bug Assignment

الملخص

In modern Open Source Software (OSS) development, developers collaborate freely over the
Internet, taking on various communication and coordination tasks. Their collective experience
and interactions are continuously recorded in software repositories such as GitHub and
Bugzilla. Unlike traditional centralized teams, online developer communities organize
dynamically; they are globally distributed across different time zones, rely on electronic
communication, and participate in projects without rigid, top-down team staffing.
Consequently, project coordinators face a significant challenge in effectively assigning
issues—such as bug fixes or version upgrades—to the most suitable developers based on their
actual experience and competency.
To address this challenge, this paper proposes a hybrid recommendation framework that
integrates Social Network Analysis (SNA) with deep learning techniques. By constructing
Developer Social Networks (DSNs) from interaction data (e.g., comments and pull requests),
we extract critical structural features, including centrality and core-periphery status. Our
methodology was rigorously evaluated using the recently released BugsRepo (2025)
benchmark.
Experimental results demonstrate that incorporating these social features—specifically
PageRank and centrality metrics—into our hybrid deep learning architecture (SNA-DeepRec)
significantly enhances recommendation performance. The proposed model achieves an
outstanding Recall of 94.07%, ensuring the consistent identification of highly capable
developers. Furthermore, it delivers a robust Accuracy of 89.35% and a Precision of 85.95%.
This exceptionally high recall effectively minimizes "bug tossing" events by accurately
pinpointing developers who are not only technically aligned with the issue's domain but also
socially central to the project's collaborative network.