Deep Hybrid Recommender System

Abstract

With the spread of e-commerce platforms in recent years, recommendation systems have become quite popular. Traditional recommendation systems are mostly based on useritem interactions. However, recommendation systems which are based on only user-item interactions are often underperforming due to the data sparsity problem. Therefore, beyond user-item interactions, the rich side information of the product or user is a notable source for improving recommendation quality. For many years, artificial neural networks have been used in many computer science fields and have gained popularity in recommendation systems in recent years. In this study, two different deep hybrid learning architectures are presented. Thanks to the feed forward neural network we use in our architectures, performance in learning the nonlinear, complex relationship between useritem interactions is increased. By adding side information to the collaborative filtering process, solutions are provided for the cold start and data sparsity problems. By making use of the strengths of deep learning and side information, it has been ensured that the constraints of collaborative and content-based methods are mitigated and the recommendation performance is increased. The success of the developed method has been compared with other studies in this field.