Efficiency and precision trade-offs in graph summary algorithms

Abstract

In many applications, it is convenient to substitute a large data graph with a smaller homomorphic graph. This paper investigates approaches for summarising massive data graphs. In general, massive data graphs are processed using a shared-nothing infrastructure such as MapReduce. However, accurate graph summarisation algorithms are suboptimal for this kind of environment as they require multiple iterations over the data graph. We investigate approximate graph summarisation algorithms that are efficient to compute in a shared-nothing infrastructure. We define a quality assessment model of a summary with regards to a gold standard summary. We evaluate over several datasets the trade-offs between efficiency and precision of the algorithms. With regards to an application, experiments highlight the need to trade-off the precision and volume of a graph summary with the complexity of a summarisation technique.