Abstract:

The opinion leader’s influence on public opinion is different under different theme communities. In order to find out the opinion leader quickly and accurately in the social network, an opinion leader mining method for the theme community is proposed. First of all, according to the proposed Interest Latent Dirichlet Allocation（I-LDA） subject model, the theme distribution with stronger ability of topic expression is obtained, and the topic similarity of the adjacent users is calculated on this basis. Then, the theme communities are divided by a multi-label equilibrium association based on the similarity of the topic, so that a user of the similar degree is divided into the same subject community, thereby further improving the accuracy and the rationality of the division of communities. With regard to the mining of the opinion leader, it proposes a Quickly-Mining Opinion Leader Algorithm（QMOLA）, which is to filter out the candidate of the opinion leader in the theme community through the structural features, and then to tap the opinion leader in the theme community by combining the characteristics of the communication and the emotional characteristics. The experimental results show that QMOLA has an obvious advantage in the mining efficiency with respect to the traditional opinion leader’s mining method, and the excavated opinion leader has higher coverage and approval rating.

Key words: opinion leader, theme community, social network, emotional characteristics, approval rating

摘要：

意见领袖在不同的主题社团下对舆情的传播影响力是不同的，为了在社交网络中快速准确挖掘出意见领袖，提出一种面向主题社团的意见领袖挖掘方法。根据提出的兴趣隐含狄利克雷分布（Interest Latent Dirichlet Allocation，I-LDA）主题模型得到主题表达能力更强的主题分布，并在此基础上计算相邻用户的主题相似度。采用基于主题相似度的多标签均衡社团划分算法划分主题社团，使相似度大的用户被划分到相同的主题社团中，由此进一步提升社团划分的准确性与合理性。对于意见领袖的挖掘，提出一种快速意见领袖挖掘算法（Quickly-Ming Opinion Leader Algorithm，QMOLA），先通过结构特征筛选出主题社团中的意见领袖候选人，再结合传播特征和情感特征挖掘主题社团中的意见领袖。对比实验结果表明，QMOLA相对于传统的意见领袖挖掘方法在挖掘效率上具有明显的优势，而且挖掘出的意见领袖具有更高的覆盖率和支持率。

关键词: 意见领袖, 主题社团, 社交网络, 情感特征, 支持率