Network data has become ubiquitous. Communication networks, social networks, and the World Wide Web are becoming increasingly important to our day-to-day life. Moreover, networks can be defined implicitly by certain structured data sources, such as images and text. We are often interested in inferring hidden attributes (i.e., labels) about network data, such as whether a Facebook user will adopt a product, or whether a pixel in an image is part of the foreground, background, or some specific object. Intuitively, the network should help guide this process. For instance, observations and inference about someone’s Facebook friends should play a role in determining their adoption probability. This type of joint reasoning about label correlations in network data is often referred to as collective classification.