Web Document Clustering Using Threshold Selection Partitioning.
Minoru Sasaki, Hiroyuki Shinnou · NTCIR · 2004
Clustering techniques have been applied to categorize documents on World Wide Web. In previous research, PDDP (Principal Direction Divisive Partitioning) is a well-known clustering algorithm. PDDP algorithm employs top-down and unsupervised clustering based on the principal component analysis and splits documents into two sets using a plane perpendicular to the maximum principal direction passing through the centroid of the documents. However, in case that the distribution of documents is biased, this algorithm difficu lt to split into two subsets accurately. In this paper, we propose a new clustering algorithm that improved the separation of the two sets considerably. We give experimentalresults using our clustering algorithm on the NTCIR-4 Web task.