Mark Last

In this paper we describe work relating to clustering of document collections. We compare the conventional vector-model approach using cosine similarity and Euclidean distance to a novel method we have developed for clustering graph-based data with the standard k-means algorithm. The proposed method is evaluated using five different graph distance measures under three clustering performance indices. The experiments are performed on two separate document collections. The results show the graph-based approach performs as well as vector-based methods or even better when using normalized graph distance measures.