Automated Essay Scoring using Ontology Generator and Natural Language Processing with Question Generator based on Blooms Taxonomy’s Cognitive Level

Abstract

Essay writing examination is commonly used learning activity in all levels of education and disciplines. It is advantageous in evaluating the student’s learning outcomes because it gives them the chance to exhibit their knowledge and skills freely. For these reasons, a lot of researchers turned their interest in Automated essay scoring (AES) is one of the most remarkable innovations in text mining using Natural Language Processing and Machine learning algorithms. The purpose of this study is to develop an automated essay scoring that uses ontology and Natural Language Processing. Different learning algorithms showed agreeing prediction outcomes but still regression algorithm with the proper features incorporated with it may produce more accurate essay score. This study aims to increase the accuracy, reliability and validity of the AES by implementing the Gradient ridge regression with the domain ontology and other features. Linear regression, linear lasso regression and ridge regression were also used in conjunction with the different features that was extracted. The different features extracted are the domain concepts, average word length, orthography (spelling mistakes), grammar and sentiment score. The first dataset used is the ASAP dataset from Kaggle website is used to train and test different machine learning algorithms that is consist of linear regression, linear lasso regression, ridge regression and gradient boosting regression together with the different features identified. The second dataset used is the one extracted from the student’s essay exam in Human Computer Interaction course. The results show that the Gradient Boosting Regression has the highest variance and kappa scores. However, we can tell that there are similarities when it comes to performances for Linear, Ridge and Lasso regressions due to the dataset used which is ASAP. Furthermore, the results were evaluated using Cohen Weighted Kappa (CWA) score and compared the agreement between the human raters. The CWA result is 0.659 that can be interpreted as Strong level of agreement between the Human Grader and the automated essay score. Therefore, the proposed AES has 64-81% reliability level.