Text Categorization via Ellipsoid Separation

Download

Microsoft Word Download (51Kb)

Description/Abstract

The problem of document classification based on their semantic content (text categorization) arises when the documents from some set have to be ranked according to their relevance to some usually predefined set of topics (i.e. web search, classification of news articles based on their dealing with business topics). In this work we are going to present a new batch learning algorithm for text classification. Our method applies non-linear ellipsoid separation to the vector space representation of text documents representations. We use `bag of words' vector representation of text documents, and maximal separation ratio method for pattern separation via ellipsoids [2] and kernel GSK algorithm [1] for feature extraction. Therefore we utilize maximization of the separation ratio and approximation of latent semantic feature extraction. We present some preliminary results which indicate high potential for the given approach.