Weighted combination of classifiers for word sense disambiguation based on dempster-shafer theory

Abstract

This paper proposes a framework for weighted combination of classifiers for word sense disambiguation based on Dempster-Shafer theory of evidence. First, by taking the confidence of individual classifiers into account, weighted com- bination of individual classifiers corresponding to distinct repre- sentations of context of a polysemous word is formulated based on Dempster's rule of combination. Second, the developed model is improved by an adaptive strategy of weight determination and a ranking-and-combination scheme. The experiment conducted for four polysemous words, namely interest, line, serve ,a ndhard, shows significantly better results in comparison with previous studies on the same datasets.