Recommender systems have been widely used in areas like news, movies, musics and advertisements. Two types of machine learning based algorithms, collaborative filtering and content-based filtering, are commonly used and demonstrated good performance. However, those algorithms are limited at capturing high-order nonlinear interactions between users and items. Because of the growing computing power and capability to acquire personalized data, many deep learning methods are proposed. This presentation will review algorithms like matrix factorization for collaborative filtering and content-based filtering and generalize them to deep learning. Similar to factorized matrix, user embeddings and item embeddings are computed in deep learning. We will them introduce some deep learning architectures used in industry.