Cluster Analysis

Cluster analysis is a set of statistical techniques the aim of which is to detect groups of objects with two complementary features:

A - High internal (within cluster) homogeneity;

B - High external (between cluster) heterogeneity.

In statistical language, the characteristics "A" and "B" respectively correspond to the within and between cluster variance.

In general, there are two kinds of Cluster Analysis techniques:

• Hierarchical methods, whose algorithms rebuild the whole hierarchy of the objects under analysis (the so called "tree"), whether in an ascending order or in a descending order;

• Partitioning methods, where the user defines beforehand the cluster numbers in which the set of objects under analysis is divided.

T-LAB uses both types of algorithms.

In particular:

· the Co-Word Analysis option uses a hierarchical method;
· the Cluster Analysis option allows the use of three different methods: one hierarchical and two partitional (K-means and Kohonen Maps);
· the Thematic Analysis of Elementary Contexts and Thematic Document Classification options use a bisecting K-means algorithm .

Some of the publications quoted in the Bibliography provide further information on the general aspects of the various methods (Bolasco S., 1999; Lebart L., A. Morineau, M. Piron, 1995), the specific aspects relating to the Kohonen Maps (Kohonen T., 1989) and the bisecting K-means method (Steinbach, M., G. Karypis, V. Kumar, 2000; Savaresi S.M., D.L. Boley, 2001).