PROCEEDINGS OF THE 21ST ANNUAL CONFERENCE ON RESEARCH IN UNDERGRADUATE MATHEMATICS EDUCATION
2018
San Diego, California
Using Machine Learning Algorithms to Categorize Free Responses to Calculus Questions
Page: 1420
Researchers in various science disciplines have begun exploring use of machine learning algorithms to categorize students’ answers to constructed-response tasks, achieving inter-rater reliability on par with that between expert raters. We report on a proof-of-concept experiment in which we categorized student responses to conceptually-focused tasks on a calculus final exam. Our results were only modestly successful, but promising. We identify ways in which responses to mathematics tasks are uniquely challenging for these algorithms, and ways in which the algorithms’ performance on mathematics tasks can be improved.