Title: Experimental analysis of impact of term weighting schemes on cluster quality

Authors: G. Hannah Grace; Kalyani Desikan

Addresses: Department of Mathematics, School of Advanced Sciences, VIT University, Chennai, India ' Department of Mathematics, School of Advanced Sciences, VIT University, Chennai, India

Abstract: Term weighting schemes are used to identify the importance of each term with respect to a collection and assign weights to them accordingly. Document clustering uses these term weights to compare the similarity between documents. Several term weighting schemes are in use today, but none of them are specific to the clustering algorithms. The term frequency-based clustering techniques consider the documents as a bag of words while ignoring the relationship between the words. So, in this paper we focus our analysis on different term weighting schemes such as term frequency (tf), term frequency-inverse document frequency (tfidf), automatic text categorisation (ATC) without normalisation and ATC-inverse document frequency (ATCidf). In this paper, we have used the clustering tool CLUTO to experimentally study the impact of term weighting schemes on the quality of the clustering solution obtained by applying the Repeated Bisection Partitional Algorithm.

Keywords: document clustering; term weighting scheme; cluster quality; criterion functions; entropy; purity.

DOI: 10.1504/IJAIP.2018.089503

International Journal of Advanced Intelligence Paradigms, 2018 Vol.10 No.1/2, pp.178 - 193

Received: 13 Apr 2016
Accepted: 03 Oct 2016
Published online: 29 Jan 2018 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article