PPRviz: Effective and Efficient Graph Visualization based on Personalized PageRank

Abstract

Graph visualization is an important problem that finds applications in various domains, e.g., social network analysis, traffic planning, and bioinformatics. Existing solutions for graph visualization, however, fail to scale to large graphs with millions of nodes, as they either provide inferior visualization results or incur significant computational cost. To address the deficiencies of prior works, we propose PPRviz, a multi-level visualization method for large graphs. Lying in the core of PPRviz is a new measure of graph node distance, PDist, that is specifically designed for visualization. In particular, PDist is formulated based on personalized PageRank, and it provides non-trivial theoretical guarantees for two well-adopted aesthetic measures. We present efficient algorithms for estimating PDist with provable accuracy and time complexity, while incurring small preprocessing costs. Extensive experiments show that PPRviz significantly outperforms 13 state-of-the-art competitors on 12 realworld graphs in terms of both effectiveness and efficiency, and that PPRviz provides interactive visualizations within one second on billion-edge graphs.