Abstract

In this paper, we applied a novel learn-ing algorithm, namely, Deep Belief Net-works (DBN) to word sense disambigua-tion (WSD). DBN is a probabilistic gen-erative model composed of multiple lay-ers of hidden units. DBN uses Re-stricted Boltzmann Machine (RBM) to greedily train layer by layer as a pre-training. Then, a separate fine tuning step is employed to improve the discrim-inative power. We compared DBN with various state-of-the-art supervised learn-ing algorithms in WSD such as Support