Visual Anomaly Detection in Temporal Knowledge Graphs

This paper addresses the visualization challenges posed by Mini Challenge 3 of the VAST Challenge 2024, which involves detecting illegal fishing activities within a dynamic network of companies and individuals. The task requires effective anomaly detection in a timedependent knowledge graph, a scenario where conventional graph visualization tools often fall short due to their limited ability to integrate temporal data and the undefined nature of the anomalies. We demonstrate how to overcome these challenges through wellcrafted views implemented in standard software libraries. Our approach involves decomposing the time-dependent knowledge graph into separate time and structure components, as well as providing data-driven guidance for identifying anomalies. These components are then interconnected through extensive interactivity, enabling exploration of anomalies in a complex, temporally evolving network. The source code and a demonstration video are publicly available at github.com/MaAllma/Temporal_Knowledge_Graph_Analysis.