Scalable Information Flow Mining in Networks.

The problem of understanding user activities and their patterns of communication is extremely important in social and collaboration networks. This can be achieved by tracking the dominant content flow trends and their interactions between users in the network. Our approach tracks all possible paths of information flow using its network structure, content propagated and the time of propagation. We also show that the complexity class of this problem is #P-complete. Because most social networks have many activities and interactions, it is inevitable the proposed method will be computationally intensive. Therefore, we propose an efficient method for mining information flow patterns, especially in large networks, using distributed vertex-centric computational models. We use the Gather-Apply-Scatter (GAS) paradigm to implement our approach. We experimentally show that our approach achieves over three orders of magnitude advantage over the state-of-the-art, with an increasing advantage with a greater number of cores. We also study the effectiveness of the discovered content flow patterns by using it in the context of an influence analysis application.