Topic-aware Influence Maximization On Online Social Network

In recent years,with the rapid development of the Internet,the large-scale online social networks such as Weibo and WeChat has developed fast,and the number of users who post messages or forward comments through social networks is also increasing.The large number of real-world social network dataset attract more and more researchers to participate in it.At present,the researcher's research directions on social network mainly include the following aspects,mining the model of information spread in the network,mining the node which has greater influence on the whole network,calculation of propagation probability between the two nodes,and influence maximization problem.The influence maximization is originated from the word-of-mouth effect adopted by early merchants to promote their products.In the existing studies of influence maximization,most of the research ignores the importance of the topic factors in information dissemination,and sets the activation probability between the two nodes to a fixed value.In this paper proposes a topic-aware information diffusion model.Two factors are considered when calculating the propagation probability between two nodes.One factor is how interested of a user to a given message,and the other factor is how often users contact each other.The probability distribution of users and messages is obtained by using the latent factor model from the recommendation system.The topic is treated as an latent factor in the model,and the original user-message matrix is decomposed into the product of the user-topic matrix and the topic-message matrix.The degree of interest for a user to a message is represented by the dot product of the user latent vector and the message latent vector.In solving the topic-aware influence maximization,the influence spread has the sub-modularity.The first part does not need the distribution of the message.Calculated the upper bound of the influence of each node,sorted the nodes according their influence.The second part is to mine the seed nodes for a certain message.By calculating the specific propagation probability generated by the message,the actual influence increment of the node on the current network which is largest are selected iterated to the seed sets.To further improve efficiency,we use community detection to cut the original network.In this paper,based on the proposed topic-aware information diffusion model,the topic-aware influence maximization algorithm is implemented.The effectiveness of the algorithm is verified by experience.