Graphics Reference
In-Depth Information
Returning to the Gephi product affinity graph, turn off filtering to see
communities across all time. Run the modularity statistic to compute them,
and use the partition pane to assign color to each community. Export a
snapshot of the result in the Preview tab. This image will represent the
final state of product communities, taking into account the full history of
reviews. Return to the Overview tab, toggle filtering back on, and restrict
the date range to the first part of the time period. Run the force-directed
layoutagain.Exportasecondimage,representingtheinitialstateofproduct
communities, and compare the results, as shown in
Figure 15-7
.
Comparing the images clearly reveals the nodes in the initial community at
top, which eventually migrate to neighboring communities. In this data set,
communities do not tend to change dramatically over time, but the example
demonstrates how you can use the technique to see changes in community
structure, however big or small.
Note
Notice that the pink community computed on the final graph (which
spatially does not look like a cluster) actually seems more cohesive in
earlier states. Also note that a few of the nodes in some of the spatial
clusters are of a different color. Cluster computation and clustering
layout can involve a degree of randomness and variability at the
detailed level, but at the broad-strokes level, will produce consistent
results.
