【摘要】 针对传统软件缺陷预测方法无法准确识别高风险模块的问题，提出双视角异构图软件缺陷预测(DeDVHG)方法。该方法结合开发者和代码两种视角构建异构图，提取节点的初始特征表示，并使用异构图神经网络进行节点特征学习。通过这种方法，可以更全面地捕捉软件网络中的复杂关系和信息，提升缺陷预测的准确性。实验结果表明，与现有缺陷预测方法相比，DeDVHG在预测性能上有显著提升，能够更有效地识别高风险缺陷模块，为优化软件测试资源分配提供了有力支持。更多还原

【Abstract】 In the context of industrial software development, addressing the problem that traditional software defect prediction methods fail to accurately identify high-risk modules, this study proposes a novel approach: the Defect Prediction Model by Leveraging Dual-View Heterogeneous Graphs(DeDVHG)method.This method constructs a heterogeneous graph by integrating two perspectives—developers and code—to extract initial feature representations of nodes, and employs a heterogeneous graph neural network for node feature learning.Through this approach, it can more comprehensively capture complex relationships and information within software networks, thereby enhancing the accuracy of defect prediction.Experimental results demonstrate that DeDVHG significantly outperforms existing defect prediction methods in prediction performance, enabling more effective identification of high-risk defect modules and providing robust support for optimizing software testing resource allocation.更多还原