Abstract:Although feature selection is very important during either supervised or unsupervised word sense disambiguation processing, there is no systematic study on investigating the relationship between the number of features and the performance as we know yet. This paper investigates the effect of the number of features to supervised Chinese word sense disambiguation through thousands of experiments on Semeval 2007 Multilingual Chinese-English Lexical Sample task dataset. It shows that local basic feature provides adequate information to do disambiguation and the influence of data sparseness is not as important on the performance as we think before from the number of features point of view.

The Effect of the Number of Features to Supervised Chinese Word Sense Disambiguation