Computer Studies Bldg. 209
Ph.D. Thesis Proposal
Modeling Complex Human Behavior from Locations and Interactions of Users in On-Line Social Networks
Recent research has shown that surprisingly rich models of human activity can be learned from GPS (positional) data. However, most effort to date has concentrated on modeling single individuals or statistical properties of aggregated groups of people. We, in contrast, take on the task of understanding human interactions, attempted interactions, and intentions from noisy sensor data in a fully relational multi-agent setting. We use a real-world game of capture the flag (CTF) to illustrate our approach in a well-defined domain that involves many distinct cooperative and competitive joint activities. Our evaluation shows that given a model of successfully performed multi-agent activities, along with a set of examples of failed attempts at the same activities, our system automatically learns an augmented model that is capable of recognizing success, failure, as well as goals of people's actions with high accuracy. Finally, we demonstrate that explicitly modeling unsuccessful attempts boosts performance on other important recognition tasks. This supports our hypothesis that modeling both attempts and failures in people's behavior is an integral---though often neglected---component of activity recognition.
Obtaining such encouraging results in the CTF game domain motivates our current and future work on modeling human behavior on a massive scale by learning from on-line social networks where users often reveal their location along with their posts. We specifically concentrate on publicly available Twitter data.
The main reason we argue for a statistical relational approach---as illustrated in the CTF domain---is the fact that our society is founded on the interplay of human relationships and interactions. Since every person is tightly embedded in our social structure, the vast majority of human behavior can be fully understood only in the context of actions of other---related---people. Solving the activity recognition problem would have a huge impact on virtually all aspects of our lives as being able to reason about people's behavior in an automated way is a necessary precondition for having intelligent and helpful machines that are aware of ``what is going on'' in human-machine as well as human-human relationships.
In this proposal, we first review our work on modeling the games of capture the flag. We then illustrate how it builds groundwork for attacking the general problem of modeling and reasoning about people's activities in our everyday lives, which is the focus of our current and future work.