融入结构度中心性的社交网络用户影响力评估算法

Social Network User Influence Evaluation Algorithm Integrating Structure Centrality

在社交网络中,通过追踪极少数的强影响力用户,可以实现宏观管控信息的传播过程,而用户影响力是一种无法预判的后验信息,仅能依靠有关特征来确定。因此,提出了一种融入结构度中心性的社交网络用户影响力评估(Structural-Degree-Centrality User Influence Rank,SDRank)算法来识别强影响力用户。该算法基于PageRank算法,引入了结构度中心性,结合了加入时间与平均转发数的调节因子,进而计算出用户的影响力值。相较于其他的现有算法,SDRank算法仅从用户本身的行为角度出发,不需要诸如个人标签、粉丝等存在伪造风险与缺省可能的具体信息,也不必挖掘传播内容的潜在信息,适用性更广泛。以微博用户的级联转发数据集作为实验对象,对被转发数排名Top-K用户的平均转发数等相关结果进行了可视化分析,探讨了用户转发行为在社交网络信息传播中的作用。在实验过程中,所提算法与PageRank,TrustRank算法相比,准确率、召回率和F1-measure值都有了一定的提高,验证了SDRank算法的有效性。

In social networks,the transmission process of information can be controlled macro by tracking a small number of strongly influential users,but user influence is a kind of posterior information that cannot be predicted and can only be determined by relevant characteristics.Therefore,this paper proposes a social network user influence evaluation algorithm that integrates structural degree centrality to identify users with strong influence.As an evaluation algorithm for social network user influence,SDRank is developed based on an improved PageRank algorithm,which introduces structural degree centrality,combines the re-gulatory factor of join time and average forward number,and then calculates the user’s influence.Compared to other existing algorithms,SDRank is applicable to a broader set of scenarios from a user behavior perspective,for it doesn’t require specific information(such as personal tags,fans)that have potential forgery risks or default possibilities,and doesn’t have to exploit the under-lying information of disseminated content.This paper takes the cascade forwarding dataset of Weibo users as the experimental object,makes a visual analysis of the average forwarding number of top-K users and other relevant results,and discusses the role of user forwarding behavior in information transmission in social network.During the experiment,its accuracy,recall rate and F1-measure value are greatly improved compared with PageRank and TrustRank,and the effectiveness of SDRank algorithm is verified.