Visualizing networks using adjacency matrices: Progresses and challenges

Visualizing networks has become a very important research and application topic in the recent years, due to the availability of network data through the web, but also to the need of analyzing several types of networks such as computer networks, social networks, biological networks (e.g. gene similarities or biological pathways). Until 2000, the node-link diagram was the only representation used. However, this representation suffers from many readability issues when the network becomes dense. In 2003, we showed that the adjacency matrix representation was more effective to visualize networks when they were dense. We conducted a controlled experiment comparing how users performed on 9 important low-level tasks required for reading a network. We varied the density and the size of the networks and measured the time to complete and number of errors for each condition using a node-link diagram and a matrix. We had significant results for 8 of these tasks, proving that the matrix representation was more effective for large and dense networks, except for one task: path following. Indeed, the matrix representation is not good at finding paths between vertices whereas a correctly laid-out node-link diagram makes it easy on sparse networks and sometimes possible on denser ones.