Until recently, complex phenomena, such as human behavior and disease epidemics, have been modeled only at an aggregate level and with many unrealistic assumptions. The main reason for such simplifications has been the lack of detailed data. While the situation is beginning to change with the surge of online social media and sensor networks, we additionally need the ability to reason about the data efficiently. In this talk, I will focus on our "TwitterHealth" line of work, where we mine millions of geo-tagged Twitter messages and identify users afflicted by flu-like illness. We then quantify the impact of social ties to infected people, as well as the intensity of recent co-location with them on one's likelihood of contracting the illness. We then show that these signals can be effectively leveraged to predict if and when an individual will fall ill with consistently high accuracy multiple days into the future. Unifying standard survey-based health monitoring with automated inference shows promise in improving our understanding of the emergence of global epidemics from day-to-day activities of individuals.
Adam Sadilek is postdoctoral fellow at the University of Rochester, where he concentrates on large-scale machine learning of human behavior exhibited both online and offline. In parallel with his studies, Adam has done research and development work at eBay Research Labs, Google, and Microsoft Research, where he focused on optimization and novel applications of machine learning in the context of big data.
Adam holds a PhD degree in computer science from University of Rochester (2012), and a BS from Czech Technical University in Prague (2008). He spent a year as an exchange student at Union College (2006).