Complex Network Neural Dynamics Framework for Automated Software Bug Triaging

Assigning bugs in large-scale open-source software projects is a critical yet challenging task due to the increasing volume of bug reports and their dynamic dependencies. Existing methods often fail to capture the structural and temporal aspects of bug reports, leading to reduced assignment efficiency and accuracy. To address these limitations, we propose a Complex Network Neural Dynamics-Based Bug Report Triage (CNND-BRT) framework. CNND-BRT combines BERT and TF-IDF for text feature extraction and constructs a dynamic complex network to model bug dependencies and historical repair information. By treating bug reports as nodes, developers as labels, and dependencies as edges, the framework captures both textual and structural features, enabling robust and accurate bug assignment. Extensive experiments on real datasets from Eclipse, Mozilla, and GCC demonstrate that CNND-BRT outperforms state-of-the-art methods. The model achieves Top-10 recommendation accuracies of 80.52%, 79.67%, and 79.16% on the three datasets, respectively, surpassing baseline methods such as PP-WGCN, ST-DGNN, GCBT, and NCGBT. Ablation studies confirm the effectiveness of combining BERT and TF-IDF and highlight the importance of modeling continuous-time dynamics in dynamic graphs. Despite its strong performance, CNND-BRT exhibits limitations on datasets with unique structural characteristics, such as Eclipse, where it slightly underperforms in Top-9 and Top-10 rankings. This suggests that the model’s performance may be influenced by the underlying graph structure. For future work, we plan to explore the integration of hyperbolic space to address feature embedding distortions in scale-free or hierarchical networks, further enhancing the model’s adaptability and predictive performance.