Incremental Lossless Graph Summarization

Concisely representing large-scale graphs is inevitable for efficient storage and analysis, which can be achieved by lossless graph summarization with many desirable properties. This technique yields a summary graph consisting of supernodes and superedges, and edge corrections. However, several algorithms developed for static graphs are inefficient in terms of time and space for dynamic graphs, represented as a stream of edge insertions and deletions. In this video, we present MoSSo, the first incremental algorithm for lossless summarization of fully dynamic graphs. Given a change in the input graph, MoSSo updates the output representation by moving nodes among supernodes based on novel ideas. Extensive experiments on real graphs show MoSSo processes a change in near-constant time, up to 10^7 times faster than running the fastest batch methods, summarizes graphs with up to 0.3 billion edges, requires sublinear memory during the process, and achieves compression ratios comparable with up-to-date batch methods.