The interaction between the end-user and the interface is almost always a personal interaction. However, there are many situations where the interfaces serve to connect people or bridge people and information artifacts (text, multimedia, etc.). The networks that emerge through these connections can be mined and used to improve the user experience in significant ways. For this talk, I'm going to relate some of our recent experiences in network and text mining to support novel interaction techniques ranging from scientific concept tracking to citation recommendation to social feed ranking. I'll talk about a few specific systems--Butterworth and CiteSight--which demonstrate this idea, but will also try to describe the approach that has made these successful. In particular, I'll focus on some of the unconventional sources of network data and transformations that we have been able to leverage to achieve desired interface improvements. By looking to large-scale networks to solve text problems, or looking to Web-scale text collections instead of the single end-user's behavior, we're able to improve the user experience for many different problems in dramatic ways.