Semantic Graph Compression with Hypergraphs

Can we model complex networks as hyper graphs and compress them for faster storage, transmission, and mining of data? In this paper, we propose a modeling and compression technique that consists of two phases: (i) mapping networks to hyper graphs by exploiting inherent or structural semantic features, and (ii) partitioning the resulting hyper graph such that similar nodes are grouped into a number of possibly disconnected parts. The partitioned hyper graph is then processed in order to yield more structural redundancy to increase compression. We provide empirical results that compare the proposed method to random and natural orderings of select real networks using an information-theoretic measure. When modeling networks using hyper graphs as proposed here, the potential for compactness and compression increases, as observed in our experimental evaluation. This benefits a variety of domains in a variety of ways, such as social networks, biological systems, and the need to represent these as compactly as possible for faster execution of queries. We also address questions for eventual investigation.