在微博网络中,消息的传播与意见领袖的影响力密切相关.然而,意见领袖在消息动态传播过程中所表现出的影响力难以量化衡量,这对意见领袖影响力的评估和消息传播趋势的预测带来了巨大挑战.针对这一问题,提出一种基于消息传播的微博意见领袖影响力建模与测量分析方法.在分析消息传播模式的基础上,采用动态有向图描述消息在微博网络中的传播过程;发现该过程可近似分解为由各个意见领袖所驱动的子过程,根据对意见领袖影响力属性特征的分析发现,该子过程可以由指数截断的幂律衰减函数来描述.对模型中各个参数进行估计,可以定量地衡量意见领袖在消息传播过程中的初始影响力、影响力衰减指数及其影响力持续时间等指标.结合新浪微博数据的分析结果显示:消息的传播范围与传播过程中参与传播的意见领袖的数量呈弱相关:虽然意见领袖的初始影响力与其粉丝数量的大小正相关,但影响力衰减指数的大小以及影响力持续时间的长短与粉丝数量几乎无关.最后,采用所建模型对真实微博消息的传播趋势进行预测,结果表明,所提模型能够较好地对热门消息的传播趋势进行预测,这对微博中公众舆论的控制及广告定点投放具有重要意义.
In microblog networks, the propagation of messages is closely related to the influence of opinion leaders. However, it is hard to quantitate the influence of opinion leaders in different types of messages, which bring great challenges to evaluate the influence of opinion leaders and predict the propagation trend of a message. To solve this problem, this paper proposes a method to measure and analyze opinion leaders' influence in microblog based on the dynamical process of message propagation. By studying the spread patterns of messages, dynamical directed graph is employed to model the propagation process. The propagation of a message can be decomposed into sub-propagations raised by opinion leaders, which can be described as exponential truncated power-law decay function. By estimating the parameters in the model, the initial influence, influence decay rate and influence insistency of opinion leaders can be evaluated quantitatively. Results of the experiment based on the data collected from Sina microblog, show the total retweeted messages is weakly correlated to the number of opinion leaders in the spread process. In addition, the number of followers possessed by opinion leaders is positively correlated to the power of their initial influence. However, it exhibits no correlation with the decay rate and insistence time of the influence. The efficiency and correctness of the proposed model are validated by predicting the spread trends of hot messages in actual microblog networks, which is important to network marketing and message propagation controls.