Automated Scoring Model for Semi-Open-Ended Question in Scientific Explanation

Scientific explanation is a crucial skill for analyzing data and drawing reasonable conclusions, especially in the context of semi-open-ended and open-ended questions. However, evaluating such questions requires significant effort from teachers, leading to a labor-intensive grading process influenced by class size and teacher expertise, potentially compromising the question's reliability. Consequently, students may lack valuable feedback, hindering the effective development of their skills. This paper presents our research focus on designing a system to automatically score students' responses. We leverage natural language processing techniques and machine learning algorithms to construct an automatic assessment model for scientific explanations in semi-open-ended questions. The responses were obtained from eighth-grade students in the fundamental science subject of the Thai curriculum, totaling 100 responses. Initially, Thai natural language processing is employed to preprocess the responses. Subsequently, N-Gram and word embedding techniques are utilized to represent the responses in a trainable format. Finally, the performance of various machine learning models trained with different word representation techniques is reported using F1-score. The experimental results demonstrate that a common method of word representation, combined with a simple machine learning model, can achieve high performance.