Uniform Exploration of Dynamic Attributed Graph via Embedded Community Comparison

Though dynamic graph exploration is common in visualization research, few works explore dynamic attributed graphs (DAG). The main reason is that introducing node attributes leads to diverse exploration tasks and rich graph patterns, making it challenging to propose an approach that supports them while avoiding a complex design to ensure good usability. This paper presents an approach to exploring DAGs. The core idea is to divide nodes at each time point into two groups based on graph structure and attribute values, respectively, and learn an embedding for each node to reflect its changes in community affiliations across two groups. Analysts can select two groups to generate node embeddings and explore patterns in their projection. Notably, the approach uniformly supports multiple exploration tasks, each corresponding to a unique selection of the compared groups, and common graph patterns, each with a distinctive visual effect in the projection. We designed a system to achieve the approach. It addresses the design requirements for applying the approach in practice, such as providing prompts for critical operations, ensuring design uniformity, and enabling comprehensive analysis of patterns associated with specific visual structures. Patterns found on open datasets in case studies and participants’ performance in a user study illustrate the general usability and effectiveness of the approach.