A method for Word Sense Disambiguation combining contextual semantic features
Liang Wen, Juan Li, Yaohong Jin, Yongjie Lu · 2016
Word Sense Disambiguation (WSD) methods based on supervised learning usually convert WSD to a classification problem. Traditional WSD methods based on supervised learning often only consider the word, position, part of speech and some other superficial morphological and syntactic features. However, for a certain kind of polysemous words, their different senses usually appear in characteristic contexts which imply different domain information. Traditional WSD methods based on supervised learning are problematic due to they lack the ability to find such semantic features. In this paper, we first develop a feature extraction algorithm to extract such semantic features. Then we try to integrate these semantic features to train the Maximum Entropy Classifier to disambiguate the specific kind of polysemous words. After making a set of contrast experiments, we find that the integration of these semantic features can significantly improve the disambiguation effect of such kind of polysemous words.