As social media usage permeates people's lives, an increasing portion of their daily behavior leaves digital traces to be used by researchers. Social scientists can hope to gain new insight into the previously hidden but digitally recorded aspects of our digital social lives. Beyond aggregate and individual-level studies of user behavior, the digital traces also enable scientific examination of the structure of social interaction through networks. At the same time, the large scale and networked nature of social media data pose a new set of challenges to be overcome through the development of sound methodologies. We take stock of current methodological promises and challenges in social media analysis. Community detection, a set of methods for the discovery of closely knit groups, is then presented as an intermediary step that enables application of existing traditional and network analytical approaches in a smaller setting more suited to social scientific questions. In closing, we argue that this network proximity-based clustering is often more useful for social media analysis than demographic grouping.