Visualizing complex networks by leveraging community structures.

Layout algorithms provide an intuitive way of visualizing and understanding complex networks. Complex networks such as social networks, coauthorship networks, and protein interaction networks often display community structures. Existing network visualization methods that are mostly based on force-directed algorithms do not fully exploit community structures, leading to layouts with intertwined nodes/edges or "hairball" issues, especially when the size and complexity of networks increase. This paper generalizes the force-directed framework and proposes a new method for network visualization exploiting community structures. The approach, entitled GRA (G eneralized R epulsive and A ttractive algorithm), first discovers communities using community detection mechanisms and then computes weighted repulsive and attractive forces between intra- and inter-community nodes. GRA simulates the nodes in a network as particles and moves them based on repulsive and attractive forces until convergence. The method is also extended to visualize larger-scale graphs by using detected communities to compress the original graph. To quantify the effectiveness of network visualization, an area estimation method based on a multivariate Gaussian distribution with noise tolerance is introduced. A layout with a high metric prevents the visualization from entanglement while making as much full use of the canvas space as possible. Case studies on complex networks of various types and sizes demonstrate that GRA achieves state-of-the-art performance and facilitates complex network analysis. • In our method, the force weights are adaptively calculated between nodes by taking communities impacts into consideration, facilitating to produce a much better visualization. • The method in the paper can be used in various kinds of complex network tasks, such as community visualization, informatics network visualization, and information propagation analysis, etc. • The method can be used to visualize very large networks up to 100000 nodes at a very higher performance compared to strong baselines. • We proposed a visualization metric that utilizes multivariate Gaussian distributions to estimate the areas of communities and evaluate the quality of visualization. The metric works well in evaluating the community structured visualization. [ABSTRACT FROM AUTHOR]