Application of Boosting to Chinese word sense disambiguation

AdaBoost.M1 is a well known boosting-based method for improving the accuracy of a given machine-learning algorithm. In this paper, we modify AdaBoost.M1 for Chinese word sense disambiguation. Unlike AdaBoost.M1 that adapts weights of training sets, in our modified algorithm, we provide a new method to adapt the classifiers´ weights. The base classifiers are trained on a small set of labeled examples, and then augmented by a large number of unlabeled examples. We report the results of systematic experimentation performed on a standard Chinese People Daily corpus (co-developed by Institute of Computational Linguistics (ICL) of Peking University, People´s Daily and Fujitsu Limited) of 12 thousand articles. These experiments demonstrate the advantage of our new weight adaptation method and its ability to overcome the noise factor within the unlabeled examples. We also demonstrate that the overall accuracy is improved as the number of classifiers increases.