Elevating Assessment: Advancing Short Answer Evaluation Through Embedding and Text Alignment

This paper addresses the critical challenge of automating short answer grading, an increasingly essential task in educational settings. We present a novel approach that integrates the Abydos package, which includes 171 string similarity algorithms, with a machine learning framework utilizing a Random Forest regressor. This combination enables the efficient and effective evaluation of student responses by extracting and utilizing textual features systematically and achieved a 60.43% correlation score on the Texas Data structure dataset. Our experiments demonstrate the practicality and superiority of this methodology in enhancing grading consistency and reducing educators' workload. The findings highlight significant improvements in grading automation, showcasing its potential to facilitate and transform educational assessment practices.