Graphics Reference
In-Depth Information
• Graphs with hundreds to hundreds of thousands of edges can be
processed on a local computer, whereas graphs with millions to billions
of edges require a multi-computer graph analysis approach.
• Graphs with thousands of edges can be directly visualized, whereas
larger graphs require a strategy or interface for selecting and filtering
down to smaller, more manageable subsets of the graph.
Size statistics also provide a quick validation that the graph data imported
correctly into the target analysis and visualization software. Does the
software show the same number of nodes and edges as noted when the data
was prepared?
Density
Graph density is the ratio of actual edges to the maximum possible edges.
A fully connected graph (that is, a dense graph ) will have every node
connected to every other node—for n nodes, the maximal graph will
approach n 2 edges for a directed graph (or ½ n 2 nodes for an undirected
graph). Density is a good summary statistic. A large dense graph can take a
long time to process. Visualizing dense graphs can result in cluttered graphs
with too many lines to visually decipher.
Keep the following in mind:
Dense graphs —You can use, filter, extract, and visualize dense graphs
in many ways. A market basket analysis (which is a graph of all
products purchased together) is a dense graph, because almost every
product combination is purchased given enough transactions.
Depending on the objective, a dense graph may be
• Reduced to just the top edges, for example, by filtering out edges
below a set threshold or by retaining only the top n edges for each
node.
• Visualized using techniques that can work well on fully connected
graphs (such as a chord diagram or matrix), or visualized using a
node-and-link technique, with edges removed via edge filtering.
Sparse graphs —By contrast, a sparse graph has a low number of
edges compared to the maximum number of edges. Graphs that have
the number of edges on the order of one, two, or three times the number
Search WWH ::




Custom Search