Research On Social Influence Analysis And Opinion Leader Mining In Social Networks

Along with their flourish, social networks have become important communication platforms for people to publish, get and discuss latest information. In social networks, the social influence directly affects the fast information diffusion and structure evolution in social networks. As users owning larger social influence, opinion leaders have direct or indirect impact on large numbers of users within a short time. So opinion leader mining has become a key point for solving practical issues in social networks. While existing researches about opinion leaders mining only consider partial features and ignore the topic relevance of opinion leaders. Therefore, the research about social influence analysis and further opinion leader mining in social networks is of great significance.In the first, social networks were regarded as maps, in which nodes and edges are respectively regarded as users and connections between them. Then internal attributes of users and interaction attributes between users were considered together to build a network graph model, which was used to analyze the social influence in social networks. In addtion, the social influence analysis was directly applied to finding opinion leaders in social networks, devoted to improving the accuracy of opinion leader mining.With the adoption of correlation theorys of graph theory and Complex Network, also focusing on the social influence analysis, two new algorithms for opinion leader mining were presented. One was called Topic Leader Rank, which exteneded the link-vote idea of Page Rank to the directed-weighted graph and nodes owning high vote scores were regarded as opinion leaders. The other algorithm was called Opinion Leader Rank. This algorithm combined the nature of power-law distribution of nodes’ degree and the random walk idea. Then it filtered out opinion leaders according to the random-walked probability of nodes in the graph. Experiments based on the data of Sina Micro Blog, the hottest internal social n etwork in recent years, were performed. The results showed that opinion leaders found by the two algorithms had better quality than other similar algorithms.At last, a prototype system for opinion leader mining was designed and implemented. By making use of the two new algorithms, this system can automatically find opinion leaders from the database and show some important features of the opinion leaders.