Design and Evaluation of a Recommender System for High School Courses in The Netherlands
by Marije de Heus
This thesis presents a newly developed recommender system for recommending high school courses in The Netherlands. The recommender system recommends a complete set of courses to a student, based on the choices of similar students that have already completed high school. A large historical database containing information of more than 20% of all new Dutch high school students was used for this recommender. The methodologies used are a structured literature review, interviews for requirements, design of the system and offline (with a historical dataset containing grades from tens of thousands students) and online (with on-site experiments at 4 high schools) experiments. The main findings of this report are the following:
- There is a definite need for an objective recommendation of high school courses by students and school counselors;
- The recommendations are not accurate;
- The recommendations received good reviews in the online experiment;
- The recommendations did not outperform the random recommendation in the online experiment;
- A serendipitous result: the offline tests have shown that recommenders can predict future exam grades with high accuracy.
Our recommendation to Topicus, based on these findings, is not to implement the recommender system. Instead, a broader search could be started, to find other possible solutions for the need for objective recommendations. One technique that could be explored further, is the prediction of grades for single courses. We expect that school counselors will find such a tool helpful in advicing students which courses to take.