HGIM: Influence maximization in diffusion cascades from the perspective of heterogeneous graph

When solving the problem of influence maximization (IM) in social networks accompanied by diffusion cascades, existing related methods face some problems, such as relying only on diffusion cascades while ignoring network topology, inaccurate estimation of node influence, etc. To tackle these problems, a novel method named HGIM is proposed from the perspective of heterogeneous graph. HGIM is consisted of a learning node representation process and a selecting influential nodes process. A heterogeneous propagation graph is first constructed to integrate network topology and diffusion cascades, where topological edges come from the social network, and action edges come from diffusion cascades. Moreover, an influence learning framework is constructed to learn node representation and estimate node influence. To select influential nodes, two seed selection strategies are proposed based on the estimated node influence. To evaluate the proposed method, a series of experiments are carried out on four real datasets. Experimental results confirm that the proposed method outperforms other state-of-the-art methods in solving the IM problem accompanied by diffusion cascades.