Clustering overlap information is also returned by the MDEX Engine.
The
PropertyMap object (in the cluster
Supplement) also includes the following set of
properties that provide clustering overlap
information.
|
Key Name |
Value |
|---|---|
|
Dgraph.SeeAlsoClusterOverlaps |
The name of the Term Discovery dimension from which this cluster was generated. |
|
NClusters |
A number indicating the number of clusters that were returned by the MDEX Engine. |
|
Cluster_i (where
|
The cluster overlap numbers for a given
cluster. Note that the cluster number (the
|
These properties provide a square matrix that has the cluster overlap
numbers. In the matrix,
number (I, j) is the estimated number of records
(from the records sampled from the navigation states) that are covered by both
cluster I and cluster j.
Note that from the definition it follows that diagonal numbers (I, I) have the estimated number of records covered by each particular cluster. These diagonal numbers tend to decrease, because of the way that the Cluster Discovery software sorts clusters (by decreasing estimated coverage).
This information can be used in application-specific ways, for example, by an application page that presents a graphical depiction of the clusters.
