Evaluation of Disambiguation Strategies on Biomedical Text Categorization

Abstract

A common and ordinary way of representing a text is as a Bag of its component Words BoW. This Representation suffers from the lack of sense in resulting representations ignoring all semantics that reside in the original text, instead of, the Conceptualization using background knowledge enriches document representation models. While searching polysemic term corresponding senses in semantic resources, multiple matches are detected then introduce some ambiguities in the final document representation, three strategies for Disambiguation can be used: First Concept, All Concepts and Context-Based. SenseRelate is a well-known Context-Based algorithm, which use a fixed window size and taking into consideration the distance weight on how far the terms in the context are from the target word. This may impact negatively on the yielded concepts or senses.

To overcome this problem, and therefore to enhance the process of Biomedical WSD, in this paper we propose a simple modified versions of SenseRelate algorithm named NoDistanceSenseRelate which simply ignore the distance, that is the terms in the context will have the same distance weight.

To illustrate the efficiency of both SenseRelate algorithm and NoDistanceSenseRelate one over the others methods, in this study, several experiments have been conducted using OHSUMED corpus. The obtained results using Biomedical Text Categorization system based on three machine learning models: Support Vector Machine (SVM), Naïve Bayes (NB) and Maximum Entropy (ME) show that the Context-Based methods (SenseRelate and NoDistanceSenseRelate) outperform the others ones.