Using Exponential Kernel for Semi-Supervised Word Sense Disambiguation

Abstract

Word sense disambiguation (WSD) in natural language text is a fundamental semantic understanding task at the lexical level in natural language processing (NLP) applications. Kernel methods such as support vector machine (SVM) have been successfully applied to WSD. This is mainly due to their relatively high classification accuracy as well as their ability to handle high dimensional and sparse data. A significant challenge in WSD is to reduce the need for labeled training data while maintaining an acceptable performance. In this paper, we present a semi-supervised technique using the exponential kernel for WSD. Specifically, the semantic similarities between terms are first determined with both labeled and unlabeled training data by means of a diffusion process on a graph defined by lexicon and co-occurrence information, and the exponential kernel is then constructed based on the learned semantic similarity. Finally, the SVM classifier trains a model for each class during the training phase and this model is then applied to all test examples in the test phase. The main feature of this approach is that it takes advantage of the exponential kernel to reveal the semantic similarities between terms in an unsupervised manner, which provides a kernel framework for semi-supervised learning. Experiments on several SENSEVAL benchmark data sets demonstrate the proposed approach is sound and effective.