Topical Influence User Analysis In Twitter

Along with the explosive growth of information, data, information and knowledge have generated new economic structure. At the same time, the social networks become extremely complex. Now there are many mature algorithms for social network analysis, however, many algorithms cannot be applied to large-scale data, influence user analysis algorithm is also facing the same problem.The goal of analyzing influential users is to find the highest authoritative users in social networks. Through in-depth analysis for social networks, we could reveal users’own characteristics and model the strength of relationship between users using the characteristics of social networks. However, the strength of the relationship between users can not be used as criteria for influential users. In order to find influential users, many ranking algorithms are proposed. The main mechanism of those ranking algorithms is to calculate the authority value of the nodes in social networks iteratively.For online social networks, the relationship strength between users not only depends on the degree of similarity between the users, but also depends on the frequency of interactions between them. The more frequent the interactions are, the greater the relationship strength between the users is. Interactions between users constitute the user interaction diagram in social networks. User interaction diagram could reflect the basic features of social networks.This paper focuses on the problem of Influential User Analysis (IUA) in social networks. In particular, we take Twitter as an example, in which users are publishing tweets, following interested users and interacting with others. The key challenge of IUA is how to leverage local attributes and global structure to quantify influential users. In this paper, we propose a novel framework for IUA by making use of the local attributes and the global structure. In order to effectively measure relationship strength among users with heterogeneous data, a poisson regression-based latent variable model is employed to infer relationship strength, and a coordinate ascent optimization procedure is developed for inference. By learning the relationship strength and analyzing the influential users, we introduce recommendation application on the proposed algorithm. Experimental results show that the proposed IUA algorithm outperforms existing related algorithms on finding influential users.