A scalable model for efficient information diffusion in large real world networks

Abstract

Real world networks represent the relationships and interactions of real world entities, such as individuals in a social network. They usually have millions of vertices, and are very complex as they are dynamic, real time and fast growing. Information diffusion models on such graphs have a major role in spreading information at a very large scale. Influential nodes are a small subset of nodes in a graph that could maximize the spread of information. Finding influential nodes in real graphs has applications ranging from viral marketing to vaccination strategies during epidemics. We propose a model to find a set of influential nodes from a very large real world graph effectively by taking into account both the structural features and centrality measures of the graph. To this end, the community structures within the graph are identified. We convert COPRA algorithm for finding overlapping communities to an equivalent Bulk Synchronous Parallel(BSP) algorithm for parallel processing of the graph. We then applied PageRank algorithm to the isolated communities to find influential nodes within each community. Finally the top k influential nodes are identified by applying a ranking function to the influential nodes generated from both isolated and overlapped communities. We found that by dividing the graph into communities and by finding influential nodes within each of those communities, a better set of nodes for fast information diffusion is obtained.