Content-based course recommender system for liberal arts education

Liberal Arts programs are often characterized by their open curriculum. Yet, the abundance of courses available and the highly personalized curriculum are often overwhelming for students who must select courses relevant to their academic interests and suitable to their academic background. This paper presents the course recommender system that we have developed for the Liberal Arts bachelor of the University College Maastricht, the Netherlands. It aims to complement academic advising and help students make better-informed course selections. The system recommends courses whose content best matches the student's academic interests, issues warnings for courses that are too advanced given the student's academic background and, in the latter case, suggests suitable preparatory courses. We base the course recommendations on a topic model fitted on course descriptions, and the warnings on a sparse predictive model for grade based on students' past academic performance and level of academic expertise. Preparatory courses consist of courses whose content has the best preparatory value according to the predictive model. We find that course recommendations are relevant for a wide range of academic interests present in the student population and that students found recommendations for courses at other departments especially helpful. The preparatory courses often lack coherence with the target course and need to be improved.