On-line single-pass clustering based on diffusion maps

Abstract

Due to recent advances in technology, online clustering has emerged as a challenging and interesting problem, with applications such as peer-to-peer information retrieval, and topic detection and tracking. Single-pass clustering is particularly one of the popular methods used in this field. While significant work has been done on to perform this clustering algorithm, it has not been studied in a reduced dimension space, typically in online processing scenarios. In this paper, we discuss previous work focusing on single-pass improvement, and then present a new single-pass clustering algorithm, called OSPDM (On-line Single-Pass clustering based on Diffusion Map), based on mapping the data into low-dimensional feature space. © Springer-Verlag Berlin Heidelberg 2007.