What Is It?

Explore’s Network Visualization and Structure determines how Nodes and Edges are spatialized and grouped within a network visualization.

Why Is This Important?

Network graphs are created using relationship data that designates how certain entities (aka nodes) interact with one another. The visualization and spatialization of these entities and their interactions with one another is dependent on the software you use to visualize it. 

Explore’s proprietary visualization capabilities are proven to be 10-100x faster than other network tools on the market in generating visuals and analysis for network data. Please reach out to info@virtualitics.com to receive a copy of our white paper!

How?

When a network dataset is loaded into Explore, we automate the computation of ‘structure’ (community detection, 3D spatial layout, etc.) These will be represented to users as a few different Network features:

• Spatialization - Network (1) X/Y/Z
• Community Detection - Louvain Community
• Edges / Weighting - Degree / Weighted Degree

In the GIF above, the Spotify Artists dataset is loaded in as a network dataset and these features are automatically computed and shown as data in the table. Flipping to the Network visualization, we notice that the spatialization features are automatically mapped to the X, Y, and Z axes, Louvain Community is mapped to Color, and Weighted Degree is mapped to Size. 

Additional Details

3D Spatialization

The ForceAtlas3D algorithm was developed by Virtualitics and is a proprietary improvement on the ForceAtlas2[https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0098679] algorithm. ForceAtlas3D is a force-directed layout algorithm[https://en.wikipedia.org/wiki/Force-directed_graph_drawing] in which nodes apply a repulsive force on each other while edges serve as an attractive spring force between pairs of nodes. A gravitational force is applied to the overall system to maintain a reasonable scale. These 3 forces (repulsive, attractive, and gravitational) are computed and applied iteratively until the network reaches a stable configuration. Explore’s ForceAtlas3D leverages GPU technology to get a substantial speedup in run-time.

Community Detection

Virtualitics uses a hybrid of Markov Clustering and Louvain Modularity to rapidly produce community detection results:

• Louvain Modularity - a community detection algorithm that aims to iteratively partition a network into “modules” or communities. 
• Markov Clustering - a community detection algorithm which leverages linear algebra operations to identify clusters/communities of nodes that have a high level of interconnectivity. 

The Virtualitics Community Detection algorithm uses each algorithm for what they are most effective at doing: Markov Clustering rapidly reduces the size of the network so that initial community labels can be calculated while Louvain Modularity handles the fine-tuning of the final community detection results.

Edge Sampling

Some Network Graphs have significantly more edges than nodes, in which case it is valuable to hide the view of excess edges. Explore will automatically show a subset of the most significant edges in a network while still providing users with a view of the overall network structure.