Machine learning and the Web are a technology and an application area made for each other. The Web provides machine learning with an ever-growing stream of challenging problems, and massive data to go with them: search ranking, hypertext classification, information extraction, collaborative filtering, link prediction, ad targeting, social network modeling, etc. Conversely, seemingly just about every conceivable machine learning technique has been applied to the Web. Can we make sense of this vast jungle of techniques and applications? Instead of attempting an (impossible) exhaustive survey, I will instead try to distill a unified view of the field from our experience to date. By using the language of Markov logic networks - which has most of the statistical models used on the Web as special cases - and the state-of-the-art learning and inference algorithms for it, we will be able to cover a lot of ground in a short time, understand the fundamental structure of the problems and solutions, and see how to combine them into larger systems.