Text Categorization for Vietnamese Documents

Many machine learning methods have been proposed for text categorization, but most research has applied them to English documents. Vietnamese is a different language with different features and it is not clear whether the standard methods will work on the categorization of Vietnamese documents. This paper describes morphological level document representations that are appropriate for Vietnamese text documents and investigates the effectiveness of several standard learning algorithms including Naïve Bayes, K-Nearest Neighbour (KNN) and Support Vector Machine (SVM) with four different kernel functions. The results show that it is possible to build effective and efficient classifiers for Vietnamese text categorization using our representations and the standard algorithms, and demonstrate that the performance can be improved by using infogain for feature selection and using an external dictionary for filtering the vocabulary.