PATHFINDER: Graph-Based Itemset Embedding for Learning Course Recommendation and Beyond

Jiasheng Zhang, Thai Le, Yiming Liao, Dongwon Lee

Abstract: We demonstrate a tool, named as PATHFINDER, that captures and visualizes rich latent relationships among courses as a graph, mines students' past course performance data, and recommends pathways or top-k courses most helpful to a given student, using an itemset embedding based learning model. With dedicated design for the asymmetric, non-additive and non-negative challenges specific to the problem, our model for helpfulness achieves the best performance among competing models. We demonstrate the visualization of four course relationships (e.g., mandatory, prerequisite, helpful, and top-k) in a graph. The PATHFINDER demo is publicly available at: http://140.82.60.177:8000/.