In this chapter, we covered recommender systems. We first looked at some background theory, implemented simple methods with TensorFlow, and then discussed some improvements such as the application of BPR-Opt to recommendations. These models are important to know and very useful to have when implementing the actual recommender systems.
In the second section, we tried to apply the novel techniques for building recommender systems based on Recurrent Neural Nets and LSTMs. We looked at the user's purchase history as a sequence and were able to use sequence models to make successful recommendations.
In the next chapter, we will cover Reinforcement Learning. This is one of the areas where the recent advances of Deep Learning have significantly changed the state-of-the-art: the models now are able to beat humans in many games. We will look at the advanced models that caused...