Categorical Clustering By Converting Associated Information
Dongmin Cai, Stephen S. Yau · Zenodo (CERN European Organization for Nuclear Research) · 2007
Lacking an inherent "natural" dissimilarity measure between objects in categorical dataset presents special difficulties in clustering analysis. However, each categorical attributes from a given dataset provides natural probability and information in the sense of Shannon. In this paper, we proposed a novel method which heuristically converts categorical attributes to numerical values by exploiting such associated information. We conduct an experimental study with real-life categorical dataset. The experiment demonstrates the effectiveness of our approach.