Spectral Document Clustering Algorithms with Different Data Structures.
Suely Oliveira, Sang-Cheol Seok · 2005
Spectral document clustering methods construct sparse word-document matrix W to measure the difference or similarity of documents. It may produce a dense similarity matrix S with W T × W. We presented a multilevel algorithm on S in [9]. The spectral clustering algorithms on S work well when the size of the dataset is not too big. However, the multiplication for S takes too much time even with efficient sparse multiplication. In this paper, we present a variant of the algorithm on W and investigate two algorithms with computational experiments.