Thinking Skills Level Classification of Scientific Questions Using Bidirectional LSTM

Science education with a suitable learning activity can help students enhance their thinking skills. Examination is one of the assessment tools to evaluate the student learning outcome in the domain of thinking skills. The Revised Bloom's Taxonomy, a well-known theory used to describe cognitive domains, divides thinking skills into two categories: basic and advanced thinking skills. Classifying questions according to their level of thinking abilities is an important task for teachers to design effective assessment tools. The objective of this study is to propose a model for classifying Thai language questions in science subjects. Initially, we used three algorithms: Bidirectional LSTM (BiLSTM), Naive Bayes (NB), and Support Vector Machine (SVM) for selecting Thai word tokenization algorithms. Then, we compare the model's performance using different feature sets. The combination of the question, training choice, and length of choice features with BiLSTM obtained an accuracy of 70%. Moreover, we employed part-of-speech (POS) tagging for feature selection. According to the findings, using nouns, verbs, adjectives, and adverbs enhances accuracy by 80.24%. This study shows the ability to use a model to categorize science questions to assist teachers in choosing questions that are appropriate to encourage higher-order thinking skills in students.