Attribute-Based Recurrent Neighbors for Dynamic Network Visualization

Abstract

Dynamic networks describe interactions between entities under study within a specific context over a period of time. These networks consist of nodes, which represent individual elements, and edges, which indicate interactions between those elements at each temporal instant. Within the network structure, various properties can be assigned to nodes (such as degree and centrality) and to edges (such as link strength and direction). To support the analysis of the phenomena embedded in these networks, the field of Data Visualization offers techniques that create visual representations aimed at highlighting connection patterns and assisting users in exploring these phenomena. Although widely used, such techniques often disregard node attributes, focusing solely on existing connections. However, these attributes may carry valuable information that can help in understanding interactions and even contribute to predicting future connections. To address this gap, this work extends the Recurrent Neighbors technique by incorporating node attributes into the node ordering process. The proposed approach provides a visualization strategy focused on identifying relationships between attributes and connections, aiming to enhance the understanding of factors that influence the evolution of interactions over time. The method was applied to two datasets that recorded social interactions among baboons, and our results demonstrate that ordering nodes based on their attributes improves the interpretability of dynamic networks, revealing social organization patterns that would likely go unnoticed in conventional visual representations.