Semi-Automatic Short-Answer Grading Tools for Thai Language using Natural Language Processing

Abstract

The past decade has witnessed enormous advancement in online educational resources. One noteworthy advancement has been the development of automatic learning platforms. The introduction of this new technology has raised questions about its effectiveness in aiding educators to improve the engagement of students and evaluate their achievement of learning outcomes. While the use of open-ended questions to assess learners' outcomes is valuable, the workload demanded of educators can increase considerably when open-ended questions are used in large classes. We have experimented with a semi-automatic method to help grade short open-ended questions answered in Thai language. Our method employed Keyword Matching and unsupervised document grouping. Fixed types of questions were tested using different algorithms. Keyword Matching was found to be an effective method for a relatively fixed, yet open-ended set of answers. For non-fixed types of answers, Document Clustering proved suitable. In generating grading tools, we adopted three methods: Keyword Matching; Sentence Vector Similarity Ranking; and Document Clustering with TF-IDF and K-Means. The algorithms were found to be useful for online learning and grading specific content-based answers which, in turn, may be used as a guide in directing educators who wish to elicit information to provide feedback.