Evaluating fuzzy clustering for relevance-based information access

This paper analyzes the suitability of fuzzy clustering methods for the discovery of relevant document relationships, motivated by the need for enhanced relevancebased navigation of Web-accessible resources. The performance evaluation of a modified Fuzzy c-Means algorithm is carried out, and a comparison with a traditional hard clustering technique is presented. Clustering precision and recall are defined and applied as quantitative evaluation measures of the clustering results. The experiments with various test document sets have shown that in most cases fuzzy clustering performs better than the hard k-Means algorithm and that the fuzzy membership values can be used to determine document relevance and to control the amount of information retrieved to the user.