Improving deepmatrix factorization with normalized cross entropy loss function for graph-basedmooc recommendation

Abstract

Nowadays, with the fast growth of the Internet, the useful role of learning online is getting increasingly popular. MOOC platforms such as Coursera, Edx, Udemy, etc. are attracting many students from all over the world, with thousands of courses constantly continually being opened and updated. This raises the question of how to suggest courses that learners are interested in. To tackle this problem, we apply the Deep matrix Factorization model to the course suggestion along with the improved loss function. The experiment shows that our course recommendation system achieves better NDCG for top K courses than other methods. And the loss function has improved in NDCG measurement compared to the original DMF model.