A Comparative Analysis On Term Weighting In Exam Question Classification

Term weighting in text classification plays a crucial role in determining the accuracy of text classifiers based on supervised machine-learning techniques. Similarly, its importance in question classification of exam questions based on Bloom Taxonomy is also no exception. This study aims to comprehensively compare several term-weighting schemes that are used in the question classification of exam questions. The contribution from this study will lead to an appropriate term weighting baseline in exam question classification. The dataset consisting of exam questions is domain-specific and is labeled with Bloom Taxonomy, which has been validated by pedagogy experts. Term weighting schemes, namely binary, TF, TF-IDF, E-TFIDF, and TFPOS-IDF, were compared in terms of accuracy using SVM and Naïve Bayes classifiers. The outcome of this study indicated that the TF variant (3) is more suitable for use in exam question classification than TF (2). TF-IDF outperformed TF and binary but outperformed by E-TFIDF and TFPOS-IDF. Using the SVM classifier, TFPOS-IDF outperformed E-TFIDF. It can be concluded that the term weightings, namely E-TFIDF and TFPOS-IDF, are effective in increasing the accuracy of exam question classification and are potential to be a new term weighting baseline for future work on term weightings instead of TF-IDF in exam question classification.