A Comparison of Two Novel Algorithms for Clustering Web Documents

In this paper we investigate the clustering of web document collections using two variants of the popular kmeans clustering algorithm. The first variant is the global k-means method, which computes “good” initial cluster centers deterministically rather than relying on random initialization. The second variant allows for the use of graphs as fundamental representations of data items instead of the simpler vector model. We perform experiments comparing global k-means with random initialization using both the graph-based and the vectorbased representations. Experiments are carried out on two web document collections and performance is evaluated using two clustering performance measures.