We investigate how automated, data-driven, personalized feedback in a\nlarge-scale intelligent tutoring system (ITS) improves student learning\noutcomes. We propose a machine learning approach to generate personalized\nfeedback, which takes individual needs of students into account. We utilize\nstate-of-the-art machine learning and natural language processing techniques to\nprovide the students with personalized hints, Wikipedia-based explanations, and\nmathematical hints. Our model is used in Korbit, a large-scale dialogue-based\nITS with thousands of students launched in 2019, and we demonstrate that the\npersonalized feedback leads to considerable improvement in student learning\noutcomes and in the subjective evaluation of the feedback.\n