Modeling direct and indirect influence across heterogeneous social networks

Real-world diffusion phenomena are governed by collective behaviors of individuals, and their underlying connections are not limited to single social networks but are extended to globally interconnected heterogeneous social networks. Different levels of heterogeneity of networks in such global diffusion may also reflect different diffusion processes. In this regard, we focus on uncovering mechanisms of information diffusion across different types of social networks by considering hidden interaction patterns between them. For this study, we propose dual representations of heterogeneous social networks in terms of direct and indirect influence at a macro level. Accordingly, we propose two macro-level diffusion models by extending the Bass model with a probabilistic approach. By conducting experiments on both synthetic and real datasets, we show the feasibility of the proposed models. We find that real-world news diffusion in social media can be better explained by direct than indirect diffusion between different types of social media, such as News, social networking sites (SNS), and Blog media. In addition, we investigate different diffusion patterns across topics. The topics of Politics and Disasters tend to exhibit concurrent and synchronous diffusion by direct influence across social media, leading to high relative entropy of diverse media participation. The Arts and Sports topics show strong interactions within homogeneous networks, while interactions with other social networks are unbalanced and relatively weak, which likely drives lower relative entropy. We expect that the proposed models can provide a way of interpreting strength, directionality, and direct/indirectness of influence between heterogeneous social networks at a macro level.