earticle

영어

Identifying influential spreaders is an important issue in understanding the dynamics of information diffusion in social networks. It is to find a small subset of nodes, which can spread the information or influence to the largest number of nodes. The conventional approaches consider information diffusion through the network in a coarse-grained manner, without taking into account the topical features of information content and users. However, for messages with different topics, the target influential spreaders may vary largely. In this paper, we propose to harness historical propagation data to learn the information diffusion probabilities on topic-level, based on which we use a greedy algorithm to iteratively select a set of influential nodes for a given topic. Specially, we design a three-stage algorithm named TopicRank to mine the most influential spreaders with respect to a specific topic. Given observed propagation data, we first use Latent Dirichlet Allocation (LDA) model to learn a topic mixture for each propagation message. Then, the topic-level diffusion probability of an edge is computed by exploiting the propagation actions occurred to it and the topic distribution of these propagation messages. Last, based on the learned topic-level diffusion probabilities, we apply optimized greedy algorithm CLEF to identify influential nodes with respect to a specific topic. Experimental results show that our method significantly outperforms state-of-the-art methods when used for topic-sensitive information spread maximization.