Natural Language Processing for scoring open-ended questions : a systematic review

In education, the performance of students is measured using summative tests, which often consist of a combination of open- and close-ended questions. The latter can be automatically scored for almost a century, though, open-ended questions are still scored manually, which is both time- and resource-intensive work. This required effort can be severely reduced by automating this task, whilst also ensuring unbiased grading. This study will propose the state-of-the-art on automatic scoring of open-ended questions, by performing a systematic literature review, based on the guidelines proposed by Kitchenham. First, (pre-)processing techniques will be discussed, especially evaluating semantic similarities. Then, unsupervised machine learning techniques are considered to analyze the processed data. Finally, all found techniques will be compared, to determine how the state-of-the-art system for scoring open-ended questions should look.