Dynamic Network Exploration workflow

The paper presents a visual analytics approach for exploring dynamic networks by treating network snapshots as points in high-dimensional space and projecting them to 2D for visualization. The workflow processes timestamped network activities through four main stages:

1. Discretisation creates temporal snapshots using sliding time windows with overlap to smooth temporal transitions and avoid missing patterns.
2. Vectorisation and normalization converts each snapshot into a feature vector, with optional normalization to emphasize specific patterns.
3. Dimensionality reduction projects high-dimensional vectors to 2D, preserving similarity relationships between snapshots.
4. Visualisation and interaction presents two coordinated views: a projection view showing snapshots as coloured points (revealing clusters, outliers, and trajectories) and a network view displaying detailed topology of selected snapshots.

Through interactive exploration, analysts identify stable states (uniform-coloured clusters), recurring states (multi-coloured clusters), outlier topologies (isolated points), and transitions between states. The method is demonstrated on artificial networks and real-world high-school contact patterns, revealing daily rhythms, break periods, and network evolution.