基于扩散序列与网络结构的用户表示联合学习框架  

USER REPRESENTATION JOINT LEARNING FRAMEWORK BASED ON DIFFUSION SEQUENCE AND NETWORK STRUCTURE

作　　者：肖菡 郭亚楠 宋帆 Xiao Han;Guo Yanan;Song Fan(Xuchang Electrical Vocational College,Xuchang 461000,Henan,China;School of Software and Application Science and Technology,Zhengzhou University,Zhengzhou 450000,Henan,China)

机构地区：[1]许昌电气职业学院,河南许昌461000 [2]郑州大学软件与应用科技学院,河南郑州450000

出　　处：《计算机应用与软件》2024年第10期293-303,共11页Computer Applications and Software

基　　金：河南省自然科学基金项目(162300410094)。

摘　　要：为了充分挖掘扩散信息与网络结构之间的相关性,该文提出一种基于扩散序列与网络结构的用户表示联合学习框架。通过分别定义在观测信息扩散序列和社交网络结构上的两个最大似然估计目标来学习用户表示,并且设计一种基于表示学习的多任务学习算法用于模型优化。此外,设计一个基于抽样的算法优化所提出的关节模型。在两个社交媒体数据集上的实验表明,该模型能够在扩散预测和链路预测任务上取得较好的预测效果,并且具有更好的鲁棒性。In order to fully exploit the correlation between diffusion information and network structure,a user representation joint learning framework based on diffusion sequence and network structure is proposed.Two maximum likelihood estimation objectives were defined on the observation information diffusion sequence and social network structure respectively to learn user representation,and a multi task learning algorithm based on learning representation was designed for model optimization.In addition,a sampling-based algorithm was designed to optimize the joint model.Experiments on two social media datasets show that the model can achieve better prediction results in diffusion prediction and link prediction tasks,and has better robustness.

关 键 词：扩散序列 网络结构 链路预测 表示学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]