基于关系嵌入的异质图神经网络链接预测模型  

Heterogeneous Graph Neural Network Link Prediction Model Based on Relational Embedding

作　　者：龙伟 张明蓝 韩敏[1,2] LONG Wei;ZHANG Minglan;HAN Min(School of Computing and Artificial Intelligence,Southwest Jiaotong University,Chengdu 610031,China;Manufacturing Industry Chains Collaboration and Information Support Technology Key Laboratory of Sichuan Province,Chengdu 610031,China)

机构地区：[1]西南交通大学计算机与人工智能学院,成都610031 [2]制造业产业链协同与信息化支撑技术四川省重点实验室,成都610031

出　　处：《计算机测量与控制》2025年第3期30-36,共7页Computer Measurement &Control

基　　金：国家重点研发计划资助项目(2023YFB3308600)。

摘　　要：异质图链接预测任务是一个具有挑战的任务;通过异质图神经网络可以学习异质图节点的节点表示,并基于链接端点的节点表示进行链接预测;基于元路径的异质图神经网络往往不能兼顾效率和性能,而传统的基于关系的异质图模型难以处理复杂的关系抑或不能充分学习异质图中的类型信息,因此提出了一种简单、轻量的基于关系嵌入的异质图神经网络链接预测模型LightREGNN;使用可学习的关系嵌入表征图中的异质类型信息,并采用TTPP模型结构从而缓解模型退化问题;还采用了跳跃链接,L2归一化等方法进一步提升模型性能;通过可靠的实验表明,提出的LightREGNN在异质图链接预测任务上相较于经典的基于节点表示的异质图链接预测模型有着更好的表现;平衡了模型的效率和性能,能够成为异质图链接预测任务上一个合适的候选模型。It is a challenge for predicting heterogeneous graph links.Heterogeneous graph neural networks(HGNNs)are used to learn the representations of heterogeneous graph nodes,and to predict links based on the representations of the nodes at the endpoints of those links.However,HGNNs based on meta paths often cannot adequately balance efficiency and performance,and traditional relation-based heterogeneous graph models are difficult to process complex relations or to fully learn and leverage the type information embedded in heterogeneous graphs.To address these limitations,a simple lightweight HGNN link prediction model LightREGNN based on relation embedding is proposed,which utilizes learnable relationship to embed heterogeneous type information in representation graph.The TTPP model is used to effectively alleviates the model degradation.Moreover,the model incorporates innovative strategies such as jumping links and L 2 normalization to further enhance its performance.Reliable experiment shows that the proposed LightREGNN has notable advantages over the classical heterogeneous graph models in the link prediction of heterogeneous graphs balancing the efficiency and performance of the model,making it a suitable candidate for heterogeneous graph with an emphasis on link prediction tasks.

关 键 词：图神经网络 链接预测 异质图 异质网络 关系嵌入 

分 类 号：TH133.33[机械工程—机械制造及自动化]