Research Of Informaiton Diffusion In Online Social Networks

The rapid development of social site(Twitter, Facebook etc.) adumbrate that social media has become hot topic and trend of current internet technology development. Users in social media can create kinds of relations(follow, friends etc.), which will generate kinds of different virtual online social networks. Users in networks not only release information, but also spread information by share action or forward action. So, online social networks support information release and diffusion. The research of information diffusion in online social networks can help web users to obtain useful information, help businesses to promote their products, help governments to regulate public opinion, and the application value of the research is huge.This paper takes real online social network data and information diffusion data as research objects, and construct the overall framework of research of information diffusion in online social networks. It focus on some problems in framework and carry out research about them. These problems include user interest model, information diffusion model, problem of information diffusion maximization, combination of information diffusion and the user recommendation and so on. Specifically, the main contents of this paper can be divided into four parts:In traditional information retrieval research, we usually use word vector to profile user profile and the weights of words are calculated by TF-IDF method. There are ternary relationships between users, resources and tags, while traditional word vector method can not make full use of above ternary relationships, and it also has the problem that one word has mutiple senmantics. For solving above problems, this paper proposes tag network model to profile users interest. In tag networks, nodes represent tags and edges represent relationships between tags. Both tags and edges have weights, which represent user interest strength and strength of association between interest separately. In particular, this paper also proposes an improved TF-IDF method to calculate the weights of tags. Experimental results on Movie Lens and Cite ULike datasets validate the effectiveness of our methods.Information diffusion prediction models can be applied to public opinion warning and explosive information detection, and has important research significance and application value. Most of current information diffusion prediction models mainly have two problems: the first problem is lack of ability of time related information diffusion prediction; the second problem is that training these models need too much time. For sovling above problems, this paper proposes a novel information diffusion prediction model, called as GT model. Different from past information diffusion prediction models, in GT model, nodes of networks are not passively influenced by their neighbors to perform actions any more, but are regarded as autonomous, intelligent and retional agents. Users calculate their payoffs of different choices to make strategic decisions. The GT model introduce the time-related user payoff, so that the GT model has the ability of predicting time dynamic of information diffusin process. This paper explore the global influence of users and social influence between users to accurately calculate the user payoff. Experimental results on Sina Weibo and Flickr have confirmed the rationality and effectiveness of the proposed model in terms of time dynamic prediction of information diffusion.Current studies of information diffusion maximization are carried out on unsigned social networks containing only positive relationships(e.g. friend or trust) between users. However, information diffusion maximization problem in signed social networks is still a challenging and ignored problem. If information diffusion maximization research do not distinguish polarities of relations between users and signed social networks are roughly treated as unsigned social networks, both the positive influence and negative influence will be mistakenly counted as positive influence. For solving above problem, this paper extends information diffusion maximization problem to signed social networks, and proposes polarity related information diffusion maximization problem and polarity related independent cascade diffusion model, and also proposes greedy algorithm to sovle the problem. Experimental results on Epinions and Slashdot show that proposed method in this paper is better than greedy algorithm that does not consider polarities of relations and other heuristic methods in term of solving PRIM problem.Social networks mainly have two functions: social interaction and information diffusion. Based on these two functions, users can make friends and obtain information in social networks. User recommendation can help users find suitable friends based on user’s preferences and network structure, which can enhance the social interaction function. Furthermore, user recommendation promotes creation of new relations in social networks, and speed up network evolution and change network structure, which directly impacts information diffusion. Most user recommendation methods ignore this point. To solve above problem, this paper introduces the concept of user diffusion degree and proposes the algorithm for calculating it, then combines it with traditional recommendation methods for reranking recommended links. So that recommendation algorithms can further promote information diffusion. Experimental results on Email dataset and Amazon dataset have confirmed the rationality and effectiveness of proposed user diffusion degree in this paper. This paper also proposes a usres recommendation algorithm based on hypergraph which can be combined with user diffusion degree. Experimental results on Sina Weibo dataset have demonstrated the effectiveness of our methods in terms of recommendation.