Recommender Systems Based on Generative Adversarial Networks: A Problem-Driven Perspective

Recommender systems (RS) now play a very important role in people's online life as they serve as personalized filters for users to discover relevant items from a sea of options. Due to their effectiveness, RS have been widely employed in consumer-oriented e-commerce platforms. Despite their empirical success, however, these systems are still confronted with two limitations: data noise and data sparsity. In recent years, Generative Adversarial Networks (GANs) have received a surge of interests in many fields because of their great potential to learn complex real data distribution, and a mass of research efforts also have demonstrated the capability of GANs to enhance RS by dealing with the challenging problems of RS. In general, two lines of research have been conducted, and the common ideas can be summarized as follows: (1) for the data noise issue, adversarial perturbations and adversarial sampling-based adversarial training often work as the antidote; (2) for the data sparsity issue, data augmentation implemented by capturing the distribution of real data under the Mini-max framework is the primary coping strategy. To gain a comprehensive understanding of these research efforts, we provide a retrospective of the corresponding studies and models and organize them from a problem-driven perspective. Specifically, we propose a taxonomy of these models, along with a detailed description of them and their advantages. Finally, we elaborate on several open issues and expand on current trends in the GAN-based RS.