SsAG: Summarization and sparsification of Attributed Graphs

The large sizes of real-world graphs in many domains make graph processing computationally prohibitive. A practical approach is to perform analytics on the compressed summary of the graph that retains the essential structure with bounded error. Many summarization methods have been proposed; however, these methods have three main shortcomings: i) mainly focused on minimizing the reconstruction error and often produce dense and memory extensive summaries ii) most of them do not scale to large graphs and iii) current techniques usually do not consider node attributes during summarization. Although, these attributes play an important role in drawing actionable analytics from real-world graphs. We devise a unified approach, SsAG, that is scalable, incorporates graph topology and node attributes for summarization and produces sparse summaries. We propose SsAG, a probabilistic approach to select nodes for merging by incorporating both the graph topology and node attribute information to make a summary. Furthermore, we present a sparsification approach to reduce the size of the summary with a negligible impact on reconstruction error. We also present analytical bounds on the runtime and provide an approximation guarantee on the quality of our solution. We compare SsAG with the state-of-the-art methods to show that SsAG is comparable in quality and more efficient and scalable. We further demonstrate the goodness of SsAG by accurately and efficiently answering the queries related to the graph structure and attribute information using the summary only.