Social Network Structure And Users’ Influence Analysis Based On Data Mining

In recent years, with the rapid development of Web2.0technology, people have stepped intothe era of online social networking, and research on social networks has gained wide attention.Online social networks can be regarded as the mapping and extention of traditional socialrelationship onto the Internet online service. Because the online social network is cheap, convenient,regional independent, it brings new user experience and has become widely used amonginterpersonal interactions.At the same time, along with the "big data era", data mining becomes more and more popular.Through social network analysis, we can find the similarities and differences between the varioussocial network structure, optimize the structure of network, improve network efficiency, andenhance the usage experience. And moreover, the analysis of user behaviors in social networks canfind the status of users in the network, infer users’ habits and interests, find the most influentialusers and provides precise marketing for enterprises. The theis contributions are following.Firstly, this thesis chooses the Sina micro-blog as real data sources, to retrieve detailed userinformation by API call, including users’ relationship, registration time, micro-blog contents etc.And then, by data screening and analysis, we infer the structural characterstics of Sina micro-blog,including social networks’ average degree, clustering coefficient and average path length, etc. Thoseresults verify the well-known small-world feature of typical social networks.Secondly, this article deeply investigates the influence power of individuals in Sina micro-blog.Specifically, inspired by the traditional Google PageRank algorithm which, however, only considersthe link relationship among users, we proposes a new algorithm, SNIRank (Social NetworkInfluence Rank), to infer social network users’ influence, which explicitly incorporates users’ realactivities on social network, the number of users’ micro-blog, the number of fans etc into theiterative process with non uniform distribution. The experimental results show that SNIRankperforms better than PageRank in terms of the number of users being coverred.Finally, considering social networks as well as users’ influences always vary with time, thisthesis proposes an improved TSNIRank (Time-based Social Network Influence Rank) algorithm,which further integrates the time factor into appropriately evaluating users’ influence. Theexperimental results show that the improved algorithm can effectively reduce the rankings of thoseinactive old user.The improved algorithm is more in accordance with social network characteristics and the requirement of commercial promotion.