A cluster is a collection of relevant terms, providing a grouping of Guided Search records that share these common terms.
All of the terms, which are dimension values, must come from the same dimension and must be a Term Discovery dimension. Clusters can be generated only if the Term Discovery dimension is available for refinements. This dimension cannot be hidden, and it must also be available from the navigation states for which you want clusters. Therefore, your application must have this dimension globally available, instead of having it available only when triggered by precedence rules.
The following features apply to the clusters:
The MDEX Engine performs dynamic clustering. That is, when a user navigates the clustering tree, it is re-clustered at any selection, allowing users to zoom into their data to practically any level.
There is no limit in the number of records that can be clustered.
Each cluster is represented by a list of terms, which provide to the user what is known as information scent: the user is instantly aware of what each cluster contains. The user can quickly understand the implied content of the clustered records.
All clusters are designed to maximize two metrics: coherence -- each cluster has only closely related records -- and distinctiveness -- two different clusters will have different records.
Each cluster has high recall. A match partial technique is typically used on cluster selection, maximizing the number of semantically related records that are returned.
For more information, see the Configuration Guidelines for Clustering chapter.
