MOOCs Recommender System with Siamese Neural Network

Abstract

Massive open online courses (MOOCs) are becoming a popular method of education, as they offer students a large-scale learning opportunity. However, the variety of MOOC courses and their frequent changes make it more difficult for students to identify relevant new information. To pique students' attention, a recommendation system (RS) is used to match the learner with the best learning resources. Most research on recommender system relies mainly on the presence of explicit feedback, while this information is commonly scarce or unavailable in MOOCs. Therefore, in this paper we use implicit feedback which is gathered passively by tracking different sorts of students' behavior to model user positive and negative preferences. We propose using Siamese Neural Networks (SNNs) to extract latent representations of students and courses based on a loss function that gives observed courses a higher preference than unobserved courses. Then, users and courses similarity are determined based on new representations. Furthermore, the other challenge is recommending courses to students with little available interaction data (cold start). To solve this problem, we employ user and course content information, which aids in the creation of more accurate representations as well. We analyze the proposed model on a real dataset obtained from XuetangX-one of China's largest MOOCs-. Experiment results show that the proposed algorithm outperforms numerous baseline algorithms.