A Trust-aware Neural Collaborative Filtering for E-learning Recommendation

Social networks can provide massive quantities of information for communication among users and e-learning communities, and the trust relationships can been employed to reveal users' preferences for improving the performance of e-learning recommendation that aim to mitigate information overload and provide users with the most attractive and relevant learning resources. However, the data sparsity problem degrades recommending performance significantly. To address this problem, a novel trust-aware neural collaborative filtering model is proposed for exploiting multi-sourced information (resource content, user rating and social trust) to predict ratings in e-learning environment. We first ties deep neural network and collaborative topic regression together, to perform users and resources latent factors learning from resource content information and users rating data. Then, we incorporate social trust into rating prediction in our model, in which users' decisions regarding ratings are affected by their preferences and the favors of their trusted friends. In addition, an approach to calculating the maximum a posteriori estimates (MAP) is proposed to learn model parameters. Empirical experiments using two real-world datasets are conducted to evaluate the performance of our model. The results indicate that the proposed model has better accuracy and robustness than other methods for making recommendations in e-learning environment.