微博中基于多关系网络的话题层次影响力分析

Topical Influence Analysis Based on the Multi-Relational Network in Microblogs

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.

In microblogs contexts like Twitter, the number of content producers can easily reach tens of thousands and a large number of users participate in the discussion of the topic, for any given topic. While this large number can generate notable diversity and not all users are equally influential, it also makes finding the true influencers, those generally rated as interesting and authoritative on a given topic, challenging. In this paper, the influence of users is measured by random walks of multi- relational data in microblogs： repost, reply, copy, and read. As the uncertainty of copy and read, a new method is proposed to determine transition probabilities of uncertain relational networks. Moreover, the combined random walk is proposed for multi-relational influence network, considering both of the transition probabilities between the intra and inter of the network. Finally, influencers are classified into two types： multi-topical influencers and single-topical influencers. Experiments are conducted on a real dataset from Twitter containing about 0.26 million users and 2.7 million posts, and the results showed that the method in this paper is more effective than TwitterRank and other methods of discovering influencers. Also, the results show that the number of multi-topical influencers is far less than that of single-topical influencers, but the effect of influence is much stronger than that of single-topical influencers.