Author Login Editor-in-Chief Peer Review Editor Work Office Work

Computer Engineering ›› 2021, Vol. 47 ›› Issue (3): 117-124,130. doi: 10.19678/j.issn.1000-3428.0057206

• Artificial Intelligence and Pattern Recognition • Previous Articles     Next Articles

An Influence Propagation Model Based on Neighbor Structure

YIN Yueshuang, SUN Yanhong, LIU Yong   

  1. School of Computer Science and Technology, Heilongjiang University, Harbin 150080, China
  • Received:2020-01-14 Revised:2020-02-26 Published:2020-03-11

一种基于邻居结构的影响传播模型

尹月双, 孙艳红, 刘勇   

  1. 黑龙江大学 计算机科学与技术学院, 哈尔滨 150080
  • 作者简介:尹月双(1995-),女,硕士研究生,主研方向为数据挖掘、社交网络;孙艳红,本科生;刘勇,副教授。
  • 基金资助:
    国家自然科学基金(619721385,61602159);黑龙江省自然科学基金(F201430);哈尔滨市科技创新人才研究专项(2017RAQXJ094,2017RAQXJ131)。

Abstract: The existing studies of social network structure influence mainly focus on the interactions between users and their neighbors,ignoring the influence of different topologies formed by different relationships between several neighbors,which is called structural influence.To study the topic,this paper constructs an influence propagation model called NS-IC based on the neighbor structures.Its parameters are the influence probabilities of different neighbor structures calculated based on the idea of the independent propagation model.Then the Expectation Maximization(EM) algorithm is used for the model to learn.The experimental results on the microblog datasets show that the NS-IC model outperforms StructInf-Basic in terms of the Mean Square Error(MSE),presision and accuracy of prediction.Moreover, the results demonstrate that the influence structures with high probabilities can significantly improve the performance of predicting user reposting behavior.

Key words: social network, node influence, structural influence, propagation model, Expectation Maximization(EM) algorithm

摘要: 现有关于社交网络结构影响问题的研究多聚焦于用户与其邻居之间的相互影响,然而在若干邻居不同关联关系所形成的拓扑结构之间也会产生影响,即结构影响。对结构影响问题进行研究,构建基于邻居结构的影响传播模型NS-IC。根据独立传播模型思想计算不同邻居结构的影响概率作为模型参数,并通过期望最大化算法进行学习。在微博数据集上的实验结果表明,NS-IC模型预测的均方误差、精度和准确率均优于StructInf-Basic方法,同时表明高概率的影响结构能够显著改善用户转发行为的预测效果。

关键词: 社交网络, 节点影响, 结构影响, 传播模型, 期望最大化算法

CLC Number: