| Abstrakt: |
Recommender systems (RSs) are prominent tools massively used in different fields of social life, e-commerce, and online platforms. The use of machine learning techniques to build RSs gives good results, but it cannot satisfy all user's requirements nowadays. The exponential growth of available data integrating social networks, and contextual and temporal properties, greatly complicate the control of user's request using traditional machine learning and decision support systems, as these techniques fail to handle massive multimedia data sources, and cannot effectively capture nonlinear relationships between users and items. Moreover, they dismiss context, social relationships, and trustworthiness. Currently, deep learning techniques are successfully used in almost artificial intelligence fields including RSs. Latest research proves that deep learning-based RSs yield promising results and outperforms traditional machine learning techniques. This paper provides a comprehensive overview of recent advances in deep learning-based RSs. We deeply analyse the challenges of these systems, and how recent research works address these challenges. Furthermore, we address the strengths and weaknesses of existing approaches in order to offer an exhaustive view for new researchers to develop new ideas when tackling the issue of RSs. |
