Summary of Key Concepts

  1. Clustering can be done for terms and/or documents, if the collection has the right characteristics (which can be tested for).

  2. Reasonably fast and space efficient clustering algorithms exist, and today's computer systems can run them on most common collections.

  3. Selection of the proper clustering algorithm for a collection can probably be based on findings of research mentioned. Some algorithms can refine such a clustering, if some particular operation needs to be supported specially, such as for frequent runs.

  4. Improvements in browsing, retrieval efficiency, and in some cases retrieval effectiveness, can result from clustering.


fox@cs.vt.edu
Thu Oct 27 06:33:06 EDT 1994