Mark Last

In this paper we compare the performance of several popular clustering algorithms, including k-means, fuzzy c-means, hierarchical agglomerative, and graph partitioning. The novelty of this work is that the objects to be clustered are represented by graphs rather than the usual case of numeric feature vectors. We apply these techniques to web documents, which are represented by graphs instead of vectors, in order to perform web document clustering. Web documents are structured information sources and thus appropriate for modeling by graphs. We will examine the performance of each clustering algorithm when the web documents are represented as both graphs and vectors. This will allow us to investigate the applicability of each algorithm to the problem of web document clustering.