An effective graph summarization and compression technique for a large-scaled graph

Abstract

Graphs are widely used in various applications, and their size is becoming larger over the passage of time. It is necessary to reduce their size to minimize main memory needs and to save the storage space on disk. For these purposes, graph summarization and compression approaches have been studied in various existing studies to reduce the size of a large graph. Graph summarization aggregates nodes having similar structural properties to represent a graph with reduced main memory requirements. Whereas graph compression applies various encoding techniques so that the resultant graph needs lesser storage space on disk. Considering usefulness of both the paradigms, we propose to obtain best of the both worlds by combining summarization and compression approaches. Hence, we present a greedy-based algorithm that greatly reduces the size of a large graph by applying both the compression and summarization. We also propose a novel cost model for calculating the compression ratio considering both the compression and summarization strategies. The algorithm uses the proposed cost model to determine whether to perform one or both of them in every iteration. Through comprehensive experiments on real-world datasets, we show that our proposed algorithm achieves a better compression ratio than only applying summarization approaches by up to 16%.