Graph Visualisation

Drawing a network is harder than storing one. The default result of throwing a graph at a layout engine is a hairball that shows the data exists and nothing else. Most of the craft is in picking a layout that matches the shape of the data, and often in noticing the data is not really a network.

Desktop

• Gephi - the usual starting point. Interactive exploration, several automatic layout algorithms, handles large graphs and many input formats.
• Cytoscape - more platform than tool, with a Java plugin ecosystem. Created at the Institute for Systems Biology in Seattle in 2002 for molecular interaction networks, and now agnostic about what the nodes and edges mean.

In the browser

• sigma.js - dedicated to graph drawing and nothing else. Give it JSON, get Canvas or WebGL, and it reads what Neo4j and Gephi export.
• Cytoscape.js - the Cytoscape model as a JavaScript library.

When it is not really a network

Check first whether the data has interconnections at all. If it is a tree or a strict hierarchy, a tree layout is faster to compute and far easier to read than any force-directed graph.

• D3 tree layouts - and circle packing when nesting matters more than branching.
• Sankey diagrams - when the weight of a connection matters more than the connection: how much flows between nodes rather than what links to what.

Analysis rather than drawing

• NetworkX - the Python library for graph analysis. Community detection, centrality, path finding. Frequently the right answer when the question is about the graph rather than about the picture.

See Graph Databases for the databases themselves.