Reducing Workload in Short Answer Grading Using Machine Learning

Abstract Machine learning methods can be used to reduce the manual workload in exam grading, making it possible for teachers spend more time on other tasks. However, when comes grading exams, fully eliminating work is not yet even with very accurate automated as any mistakes could have significant consequences students. Here, evaluation of an approach therefore extended from measuring relation accuracy also overall required correctly grade a full exam, and without support machine learning. The was performed during introductory computer science course over 400 consisted 64 questions relatively short answers two-step applied. First, subset manually graded next training data models classifying remaining answers. A number different strategies how select which include were evaluated. spent actions measured along reduction effort using clustering scoring. Compared substantial—between 64% 74%—even complete review all classifier output ensure fair grading.