Deep Generative Session-Based Recommender System

The inherent structural sequences in sessions and the mutual influence of complex variables in different time steps make deep generative models effective solutions for a session-based recommender system (SBRS). In addition, in real-world scenarios, users usually only select a limited number of items, and their interactions in response to items are very sparse. Deep generative models that produce more training samples can help reduce the data sparsity problem. To this end, we discuss different deep generative models in SBRS in this chapter, such as autoencoders (AE), generative adversarial networks (GAN), and flow-based models (FBM).