Feature Selection for Bloom’s Question Classification in Thai Language

Abstract

Bloom’s taxonomy cognitive domain is a list of knowledge and the development of intellectual skills words. It is widely used in an assessment. Currently, in Thai language, teacher identifies Bloom’s taxonomy cognitive level manually, which is a tedious and a time-consuming task. This study presents automatic natural language question classification in Thai, feature selection is focused. Several previous works have been studied to fulfill Bloom’s taxonomy cognitive domain; however, those works cannot apply to Thai language due to the language characteristic. This study shows that verb, adverb, adjective, conjunction, and question tag should be selected as features in Thai’s exam classification. The dataset has been collected from a number of websites on Bloom’s cognitive domain literature. Each question is processed through cleaning data, word segmentation, part-of-speech tagging, and feature selection. After that selected feature, 70% of data set is used for training into a model. Four different classifier models, namely, Naive Bayes, decision tree, multilayer perceptron, and support vector machine are used to show the effects of the proposed feature selection technique. The results from the testing data (30% of data set) show that the proposed technique with support vector machine gives the good value of accuracy, precision, and recall, which is 71.2%, 72.2%, and 71.2%, respectively.