A visualization centered on pairwise interactions:

• A flight network
• A network of cooperative interactions between dolphins
• A network of aggressive interactions between fish
• A network of proteins and biochemical pathways
• A hierarchical diagram of relationships

Interactors are generally indicated by nodes, and interactions by edges

• Weighted networks have different importance to the edges
• In directional networks, edges have direction
• In multipartite networks, nodes are divided into two types that interact only with each other
  • almost always bipartite
• Multimodal networks show different types of links

• Relationships between particular nodes
• Patterns of clustering
• Paths and distances (six degrees of separation)
• Relationship between network structure and other variables

• May reflect the network structure
  • Usually by using force-directed algorithms (hairball)
  • Or hierarchical clustering (dendrogram)
• Or something else about the data
  • most often physical location
  • Or a logical pattern
  • or categories (hive plot)

• Size can be used for importance (should have a scaleable, physical interpretation)
  • Can be good for things about role in the network (e.g. betweenness, centrality)
• Color can be used for quantitative or qualitative attributes
  • Choose a color scale that reflects this role
• Shape can be used for qualitative attributes
  • It is possible to be creative about shape; e.g., use png icons

Strongly constrained; they’re basically lines, and the nodes are drawn first

• Use width to indicate edge “weight” (nothing else)

• Use color to indicate edge type for a multi-modal graph

• Use arrows to indicate direction on a directional graph

• Curvature can be used to show edges more clearly
  • Rarely, to carry information

• Nodes might be effectively interchangeable

• They may carry “type” information (male, female)

• We may be interested in specific names
  • These can be shown for smallish networks
  • Or added with tooltips in dynamic graphics

• Don’t drop weights without a good reason
  • consider carefully how weights are scaled
• Similarly for dropping specific edges

• Level of detail and messages will change a lot depending on how many nodes you have
  • Very big networks may show only the pattern of node clustering
• If you have many nodes, but only a few important ones, it may be possible to follow the above edge principles
  • Rescale node size to improve visualization
• Communicate clearly about rescaling

• Layout is deep, and there is no one best answer

• Try weights in your force-directed algorithm
  • Layout may better reflect data
  • Layout may look ugly

• Overview
• igraph
• tidygraph and ggraph are built on igraph, and may be a good starting point
• statnet offers a competing set of packages, including network

• Visualization with R from Katya
• FlowingData examples

• Locate and read media data
• Simple plots
• ggplots
• tidygraph