Dimension reduction techniques for accessing Chinese readability

Abstract

Machine learning-based techniques have been used to assess document readability in recent studies. One of the important issues of machine learning-based text classification techniques is to reduce the dimension of the document vectors. Different feature selection and feature extraction methods such as mutual information, chi-square test, information gain, PCA, and LSA are compared for assessing Chinese readability. We also compare classification techniques SVM and LDA. The experimental results indicate that the combination of chi-square feature selection method and SVM performs well.