Animating the development of Social Networks over time using a dynamic extension of multidimensional scaling

The animation of network visualizations poses technical and theoretical challenges. Rather stable patterns are required before the mental map enables a user to make inferences over time. In order to enhance stability, we developed an extension of stress-minimization with developments over time. This dynamic layouter is no longer based on linear interpolation between independent static visualizations, but change over time is used as a parameter in the optimization. Because of our focus on structural change versus stability the attention is shifted from the relational graph to the latent eigenvectors of matrices. The approach is illustrated with animations for the journal citation environments of Social Networks, the (co-)author networks in the carrying community of this journal, and the topical development using relations among its title words. Our results are also compared with animations based on PajekToSVGAnim and SoNIA.

💡 Research Summary

The paper tackles a fundamental challenge in the visualization of evolving social networks: how to animate changes over time while preserving the viewer’s mental map. Traditional animation pipelines first compute a static layout for each time slice and then interpolate between these layouts, typically using linear interpolation. This approach, however, suffers from two major drawbacks. First, even minor variations in the underlying data can cause large, abrupt movements of nodes, breaking the continuity that users rely on to track the evolution of the network. Second, focusing solely on the observable adjacency matrix ignores the latent structural information that often drives meaningful change, such as shifts in community cores, emerging research topics, or the rise and fall of influential journals.

To address these issues, the authors propose a dynamic extension of multidimensional scaling (MDS) that incorporates temporal smoothness directly into the stress‑minimization objective. In classic MDS, the stress function measures the discrepancy between the pairwise distances in the high‑dimensional space (derived from the adjacency or similarity matrix) and the Euclidean distances in a low‑dimensional embedding. The new formulation adds a penalty term that forces the position of each node at time t to stay close to its position at time t‑1. Formally, the overall objective is

\