Graphics Reference
In-Depth Information
Figure 15-7:
Computing communities and assigning them colors that
carry over to previous views shows how they evolved from previous
communities. Here, members of the top-most product community split into
neighboring clusters, and, surprisingly, the pink community computed in
the later view is actually more cohesive in previous views.
Transaction Graphs
In the product affinity data set used in the examples thus far, a link
represents a customer product interest, the nodes of interest are products,
and the goal with respect to time is to understand how associated product
interests changeovertime. Changeinproductinterest ismanifested visually
by the appearance and disappearance of product nodes. An understanding
ofrelatedproductinterestsisavaluablesourceofinformationformarketing
decisions. Knowing that customers interested in visualization design topics
are also interested in software development is an indication that
cross-advertising should help improve sales.
Similarly valuable would be an understanding of customer purchasing
patterns over time by customer and customer profile. For example, knowing
that customer experience with certain topics triggered increased follow-on
sales would be reason to put more marketing emphasis on those topics.
Patterns like purchasing, which involve a series of transactions over time,
would not be readily visible using the approach taken thus far.
Links in a graph indicate relationships between entities. A series of graphs
portrays the overall pattern of relationships at each point in time and can
be compared using the techniques described earlier in this chapter.
Transaction graphs
go one step deeper and articulate the series of events
within each relationship in the graph for the purposes of understanding
patterns ofbehavior. Unique approaches tovisualization and interaction are
required to support transaction graph analysis.
Clustered Transaction Analysis
The immediate challenge presented by transaction graphs is one of scale.
Many dynamic graphs already represent a Big Data problem, but given that
transaction graphs involve the addition of a whole new dimension, they
virtually
always
do. Effective visualization and navigation of transaction
graphs requires strategies to deal with scale. Hierarchical clustering and
aggregation of linked nodes is one such strategy. Chapter 14, “Big Data,”
