Reducing large graphs to small supergraphs: a unified approach

Summarizing a large graph with a much smaller graph is critical for applications like speeding up intensive graph algorithms and interactive visualization. In this paper, we propose CONditional Diversified Network Summarization (CondeNSe), a Minimum Description Length-based method that summarizes a given graph with approximate “supergraphs” conditioned on a set of diverse, predefined structural patterns. CondeNSe features a unified pattern discovery module and a set of effective summary assembly methods, including a powerful parallel approach, k-Step, that creates high-quality summaries not biased toward specific graph structures. By leveraging CondeNSe ’s ability to efficiently handle overlapping structures, we contribute a novel evaluation of seven existing clustering techniques by going beyond classic cluster quality measures. Extensive empirical evaluation on real networks in terms of compression, runtime, and summary quality shows that CondeNSe finds 30–50% more compact summaries than baselines, with up to 75–90% fewer structures and equally good node coverage.