Research On Topic-based Influence Maximization In Social Networks

With the development of Internet technology, social networks have penetrated into every aspect of people’s lives. Activities undertaken by people every day have produce amounts of information propagated among users in the social networks, and the information contains a variety of topics, there is a great impact on the users of the other users under different topic, through these users can spread the information to the greatest extent. Under this background, the topic-based influence maximization in social networks has become a hot research point. The aim is for the different topics produced in social networks, mining the most influential set of nodes under any particular topic. Then use the set to propagate information that can make the propagation of information in the social network to maximum.In this paper, the relevant theory and technology of the topic-based influence maximization has been introduced firstly, and some of the existing problems in recent years’ research have been analyzed. Then, based on the existing research, a depth research of topic-based influence maximization has token; the main research contents are as follows:(1) A topic model called user_word topic model which is suitable to mine the topic of short text in a social network is proposed to solve the problem that how to get the hidden topics and the distribution of nodes in the social network. Firstly, use the clustering algorithm to cluster the short text documents in a social network, and obtain the suitable cluster number by the sum of square error and clustering coefficient. Then intergrated the short text in each cluster to form a long text document in a document, and model the bitem generating method in each long text document according to the topic distribution of users. Then use the gibbs sampling method to infer the parameters in the model to obtain the topics and the distribution of nodes in the social network. Finally, the superiority of the proposed model is verified by the experiment in terms of topic quality, perplexity and topic differences.(2) On the basis of the above results, a topic-based influence maximization algorithm is proposed to mining the most influential set of nodes under specific topic in a social network. The algorithm consider that topic has an impact on influential nodes mining, all the nodes in a social network are preprocessed to get a subset of these nodes first; then a two-stage method is proposed for influential nodes mining on the nodes subset: the first stage mining nodes with high topic authority statically to add to the influential nodes set, the second stage use the nodes from the first stage to spread information by the topic information propagation model which is proposed in this paper, then mining nodes with biggest influence increment greedily to add to the influential nodes set. Finally, the effectiveness of the algorithm is verified by experiments.