Semantics-Aware Autoencoder

Recommender Systems are widely adopted nowadays in many services, such as e-commerce websites, content streaming platforms for both music, videos, or just news. They aim to help users to find what they look for by filtering only the relevant content to them in a personalized fashion since every user has its tastes. Over the years, several algorithms have been developed to solve the recommendation problem. Very recently, we assisted in the rise of Deep Learning, which had been able to outperform many state-of-the-art machine learning algorithms. On the other hand, even though deep learning is very effective for this problem, it is hard to explain as the model is not interpretable. In this thesis, we present SemAuto, a novel deep learning interpretable architecture that can explain its outputs, and that can be used to generate an explanation for the provided recommendation. We evaluated our semantics-aware approach with respect to other state-of-the-art algorithms to prove the recommendation's accuracy effectiveness. Furthermore, we performed an extensive A/B test with real users to evaluate the explanation we generate.