Predictive models of human behavior--and in particular recommender systems--learn from large graphs of historical interactions among users and items, in order to make personalized predictions that adapt to the needs, nuances, and preferences of individuals. After introducing the current state-of-the-art in personalized recommendation technology, in this talk I'll cover two emerging directions in recommender systems research: (1) Work that incorporates Recommender Systems with NLP techniques, such as developing personalized language models; and (2) Work that incorporates generative modeling techniques into recommender systems, including using recommender systems to facilitate the design of new products.