NtCF: Neural Trust-Aware Collaborative Filtering Toward Hierarchical Recommendation Services

It is already certificated that collaborative filtering algorithms could alleviate such data sparsity and long tail distribution problems and provide high performance in item recommendation. However, high computational complexity and insufficient samples may lead to low convergence and inaccuracy in traditional recommender approaches. In this article, a novel deep neural network-based collaborative filtering recommender engine referred to as NtCF is proposed, which resorts to a neural architecture for preference learning and user representation. With the powerful capability of neural network, NtCF is able to deep exploit interactions within social network for each user. More specifically, the trust-aware attention layer is designed to indicate the social influence to each user; furthermore, NtCF performs item clustering via k-means++ and conducts item recommendation within each generated item cluster, and accordingly, NtCF can achieve significant improvement in recommendation performance and provide hierarchical recommendation services. In practice, experimental comparison over three real-world datasets also demonstrates the superiority of NtCF in contrast to state-of-the-art recommender approaches, which can achieve high performance in top-N recommendation and provide much better user experience.