Architectural layout for a text classifier for electronic mail

Abstract

This document is summary of the investigative work done into the various components, which make up a text classification system for e-mail documents. The final intent of this research is the creation an application (TextCat) that uses machine learning techniques to classify an e-mail document into predefined semantic classes. The resulting software program performs text classification, based on distributional approximation of feature word terms, extracted from a very large corpus of labelled texts. It enables document categorisations based on a full-text of articles. Using probability theory as a formal foundation, a Machine Learning based method was developed to allow electronic mail document collections to be automatically organized at a topical level. The classification decisions are made by considering only the presence (or absence) of a small number of features (words) in each document. Various feature selection methods were investigated.