A novel centrality-based method for visual analytics of small-world networks

• Published in: Journal of visualization 2019-10, Vol.22 (5), p.973-990, Article 973
• Main Authors: Lin, Chun-Cheng, Huang, Weidong, Liu, Wan-Yu, Wu, Sheng-Feng
• Format: Article
• Language: English
• Subjects: Algorithms; Analytics; Classical and Continuum Physics; Clustering; Computer Imaging; Engineering; Engineering Fluid Dynamics; Engineering Thermodynamics; Graph theory; Heat and Mass Transfer; Information management; Intelligence (information); Nodes; Pattern Recognition and Graphics; Regular Paper; Social networks; Sociology; Vision
• ISSN: 1343-8875; 1875-8975; 1875-8975
• DOI: 10.1007/s12650-019-00582-5

Nowadays, the network data that we need to deal with and make sense of are becoming increasingly large and complex. Small-world networks are a type of complex networks whose underling graphs have small diameter, shorter average path length between nodes, and a high degree of clustering structures and can be found in a wide range of scientific fields, including social networks, sociology, computer science, business intelligence, and biology. However, conventional visualization algorithms for small-work networks lead to a uniform clump of nodes or are restricted to a tree structure, making the network structure difficult to identify and analyze. This work provides a new visual analytical method to improve the situation. Different from previous methods based on spanning trees, this method first generates a weighted planar sub-network based on the measurement of network centrality metrics. A force-directed algorithm based on node-edge repulsion is then applied to visualize this sub-network into a proper layout for better understanding of the data. Finally, the remaining links are placed back to maintain the original network’s integrity. The experimental results show that compared to previous methods, the proposed method can be more effective in differentiating clusters and revealing relationship patterns among individual nodes and clusters in the network. Furthermore, the proposed method is applied to a data of the semiconductor wafer manufacturing industry as a case study. The work shows that this new approach allows users to gain useful insights into the data. Graphic abstract