Graphics Reference
In-Depth Information
variety of other items). This type of analysis can be computed in different
ways; graph analysis is only one approach.
Note
For a detailed account of the beer and diapers story, see:
www.dssresources.com/newsletters/66.php .
Market basket graphs tend to be very dense. When aggregating data across
thousands of baskets, it is not uncommon for two unrelated items to get
purchased together a few times. Because market basket graphs are dense,
interaction is required to explore the graph and gain insight.
The example of a market basket analysis in this chapter uses anonymous
data from a hardware store to view more than 300,000 items purchased in
more than 10,000 basket purchases. The graph has approximately 8,000
products and 180,000 links. The first half of this chapter shows the various
types of interactions used to do an exploratory analysis. Gephi is used for
the screen shots, but Cytoscape or other graph software could have been
used. You can see the same graph in more detail using the MarketBasket
examples in the Supplementary Materials available on this topic's
companion website.
The initial view of the data in Gephi may look like a square with random
node layouts, as shown in Figure 6-1 . The nodes have been sized and colored
by the total revenue for that item (pink is low revenue, and deep purple is
high revenue). The edges have been colored by the frequency at which each
pair of products has been purchased.
 
Search WWH ::




Custom Search