9.2.3: Network Graphs (Node-Link Diagrams)

Network Graphs (Node-Link Diagrams)

A network graph (or network diagram) visualizes relationships between entities using nodes (points) and links (lines connecting them) [9]. Each node represents an entity (e.g. a person in a social network, a computer in a network topology, or a concept in a knowledge graph) and each link represents a relationship or connection (friendship, communication, association, etc.). Network graphs are powerful for examining connectivity patterns: they can reveal clusters of highly interconnected nodes, identify key hubs or outliers, and illustrate the overall structure (e.g. a hub-and-spoke vs. a distributed web). They are used in social network analysis, knowledge graphs, supply chain networks, fraud detection (linking entities like transactions, IP addresses, accounts), and more generally whenever pairwise relationships are fundamental to the data.

When to use a network graph: Use network graphs when the relationships between items are as important as the items themselves. If you have data that naturally forms a network (for instance, who communicates with whom, which proteins interact in a cell, which webpages link to each other), a network diagram is the natural choice to see those connections. Network graphs excel at showing clusters and communities – subsets of nodes that are densely connected internally but perhaps sparsely connected to other groups. For example, in a network graph of a company’s email communications, you might see clusters corresponding to departments (people email colleagues in their department frequently, forming tight clusters, with fewer connections between departments). They can also highlight central nodes (nodes with many connections or high importance in connecting different clusters, sometimes called hubs or with high centrality). Such visual insights are hard to gain from non-network charts.

However, network graphs can become problematic when the network is very large or dense. Beyond a certain number of nodes/links, a network graph can turn into an unreadable “hairball,” where so many links overlap that little can be discerned. Thus, network visualizations have a limited data capacity. They are best for moderate-sized networks or after filtering to a focused subset of the network. Techniques like clustering, filtering, or interactive exploration (e.g. zooming into parts of the network) are often employed to manage complexity. In terms of presentation, it’s often useful to accompany network graphs with summary metrics (e.g. number of clusters found, size of largest cluster, top hub nodes) because interpreting the graph by itself can be challenging for audiences not used to them.

From a tooling standpoint, creating network graphs can be done with libraries like D3.js, Gephi, NetworkX (Python), or even tools like Neo4j Bloom for graph databases. BI tools are not typically oriented towards network graphs (though some plugins or custom visuals exist, like network graph visuals in Power BI or Flourish for simple cases). Thus, network graphs often require more custom work or specialized software. If an interactive element is added (e.g. the ability to drag nodes or zoom into sub-networks), it can greatly enhance exploration, letting users focus on one relationship path at a time.

In summary, network graphs are the go-to for relationship data: they illuminate how entities interconnect, which ones act as bridges or hubs, and the overall topology of the network. Use them when connections matter more than individual isolated values, and when you want to visually emphasize “these things are related to those things.” But ensure to avoid overly dense network visuals or mitigate complexity by focusing on the most relevant nodes and links to tell your story.