The rise of new online platforms for capturing many aspects of our daily lives has opened the door for large-scale computational studies of nearly all facets of behavior, such as exercise, collaboration, and community dynamics. My research aims to understand and predict these kinds of human behaviors by combining techniques from natural language processing and network science to produce holistic models of people and their social interactions. The first part of this talk focuses on the challenge of inferring the demographics that describe who people are. Using the example of location inference, I show how we can efficiently and accurately learn these aspects for hundreds of millions of people across the globe. I also discuss my recent work on algorithmic bias in demographic inference and show how to mitigate inequality for the ubiquitous task of language identification. The second part of the talk shifts from individuals to their social interactions and I describe my recent work analyzing how people’s offline behavior and communication strategies change when they join online groups. I conclude by highlighting future directions in computational social science that I am excited to pursue through the combined lens of language and networks.