Attentional Autoencoder for Course Recommendation in MOOC with Course Relevance

With the rapid development of online education, a large number of students are learning in the Massive Open Online Course (MOOC) platform. To attract students’ interest, the recommendation system is applied by MOOC providers to recommend courses to students. State-of-the-art course recommendation systems still face several challenging problems: (1) It is difficult to model the nonlinear implicit feedback interaction between students and courses; (2) Text information is not contained in the network, such as the course description. To solve these problems, we propose an Attentional Manhattan Siamese Long Short Term Memory (AMSLSTM) network and based autoencoder to construct course relevance from the course description and self-attention adaptively distinguish the student’s preference level in multiple aspects. The proposed AMSLSTM learns the preference of student’s interest by considering the course relevance, and can significantly improve the recommendation accuracy. We compare the existing recommendation system baselines to evaluate our model. The experimental results show that the proposed model is effective in the real data set.