Visual aggregation of large multivariate networks with attribute-enhanced representation learning

Abstract

It is a challenging task to explore large-scaled multivariate networks, since nodes of multivariate networks always contain rich information. We propose a novel visual abstraction method to adaptively aggregate nodes and layout multivariate graphs. First, a novel attribute-enhanced graph representation learning model is proposed to transform nodes into an attribute-enhanced vectorized space. Then, we propose a multi-scale aggregation scheme to classify nodes into hierarchical categories with respect to a set of metrics such as structure closeness, attribute homogeneity and cluster account. Further, we design and implement a visualization framework enabling users to conduct attribute-aware visual abstraction, exploration, and clustering of large-scale multivariate graphs. Case studies and quantitative comparisons with three datasets verify the effectiveness of our approach in enhancing the readability of large-scaled multivariate networks.