微博中基于多关系网络的话题层次影响力分析

发布时间：2018-06-14 22:18

本文选题：话题影响力个体 + 社会网络　；参考：《计算机研究与发展》2013年10期

【摘要】：微博服务每天产生大量涉及多个话题的信息,不同用户参与话题的讨论、传播等表现出不同的影响力.为了全面度量微博中用户在话题层次上的影响力,综合考虑4种网络关系:转发关系、回复关系、复制关系、阅读关系.针对复制关系和阅读关系的不确定性,给出了网络内部转移概率计算方法;针对多关系网络,提出了基于多关系网络的随机游走模型MultiRank,分别考虑了网络内部的转移概率和不同网络之间的跳转概率.最后将影响力个体根据其影响力属性分为"多话题层次影响力个体"和"单话题层次影响力个体".真实的Twitter数据集上验证了MultiRank的有效性,实验结果表明MultiRank优于TwitterRank和其他影响力个体发现方法,同时实验结果也表明多话题层次影响力个体数目相对所有影响力个体仅占少部分,但影响效果却明显高于单话题层次影响力个体.
[Abstract]:Weibo services generate a lot of information about multiple topics every day, different users participate in the topic discussion, spread and show different influence. In order to measure the influence of users on topic level in Weibo, four kinds of network relationships are considered: forwarding relation, reply relation, copy relation and reading relationship. Aiming at the uncertainty of replication and reading relationships, this paper presents a method to calculate the internal transition probability of the network, and for multi-relational networks, MultiRank-based random walk model based on multi-relational networks is proposed. The transition probability within the network and the jump probability among different networks are considered respectively. Finally, the influence individual is divided into multi-topic influence individual and single-topic influence individual according to its influence attribute. The effectiveness of MultiRank is verified on real Twitter datasets. The experimental results show that MultiRank is superior to Twitter Rank and other influential individuals, and that the number of multi-topic influential individuals is only a small part relative to all influential individuals. But the influence effect is obviously higher than single topic level influence individual.
【作者单位】：国防科学技术大学信息系统与管理学院;国防科学技术大学信息系统工程重点实验室;国防科学技术大学计算机学院;
【基金】：国家“九七三”重点基础研究发展计划基金项目(2013CB329601,2013CB329602) 国家自然科学基金项目(60933005,9112400271331008,61302144)
【分类号】：TP393.09

【参考文献】