基于非负张量分解的动态社交网络链路预测  

Dynamic social network link prediction based on Non-negative Tensor Factorization

作　　者：杨兆鹏 袁华强 YANG Zhaopeng;YUAN Huaqiang(School of Computer Science and Technology,Dongguan University of Technology,Dongguan 523000,China)

机构地区：[1]东莞理工学院计算机科学与技术学院,广东东莞523000

出　　处：《电子设计工程》2024年第8期7-11,17,共6页Electronic Design Engineering

基　　金：国家自然科学基金(61972090)。

摘　　要：随着社交网络规模的快速增长,网络链路数据之间的发现与补全,已经是学界与工业界共同关注的一项课题。但是,传统链路分析算法多聚焦于历史链路的发现,难以准确预测社交链路中随时间变化的趋势和适应动态社交网络上的预测要求。针对以上问题,该研究提出了一种结合非负张量分解(Non-negative Tensor Factorization,NTF)和霍尔特-温特(Holt-Winters,HW)时序数据预测方法的社交网络链路预测模型。该模型使用NTF从历史图中提取节点特征,再利用HW预测方法感知节点特征随时间变化的趋势,从而实现对未来图中的链路结构的预测。结果表明该模型的AUC值相较于现有的方法提升了2.2%~12.75%,具有较好的适用性,为社交网络的链路预测问题提供了一个较好的解决方案。With the rapid growth of the social networks scale,the discovery and completion of network link data has become a topic of common concern in both academia and industry.However,traditional link analysis algorithms mostly focus on the discovery of historical links,and it is difficult to accurately predict the trend of social links over time and adapt to the forecasting requirements on dynamic social networks.Aiming at the above problems,this study proposes a social network link prediction method that combines Non-negative Tensor Factorization(NTF) and Holt-Winters(HW) time series data prediction methods.The model uses NTF to extract node features from historical graphs,and then uses the HW prediction method to perceive the trend of node features changing over time,so as to predict the link structure in future graphs.The results show that the AUC value of the model is improved by 2.2% to12.75% compared with the existing methods,which has good applicability and provides a better solution for the link prediction problem of social networks.

关 键 词：动态社交网络 链路预测 霍尔特-温特预测 非负张量分解 

分 类 号：TN915.03[电子电信—通信与信息系统]