Research

Developing Adaptive Social Robot Tutors for Children [HRI '16] [HRI '17]

We are currently working on a research project involving robots as tutors, with the goal of providing personalization within tutoring interactions. In order to build such adaptive tutoring robots, we seek to understand what types of behaviors children engage in during a learning interaction with a social robot. This project focuses on identifying salient aspects of a tutoring interaction that impact learning, and utilizing social robots to intelligently respond to these needs of children during learning. We have designed and run several HRI studies that allow children to interact with a robot tutor while completing math problems on a tablet device. The goal of these studies is to understand how various supportive behaviors that a robot tutor can employ during the interaction can be personalized to each child. Supportive behaviors include shaping help-seeking behavior, providing breaks during a long interaction, and encouraging thinking out loud during the problem-solving process.

Socially Assistive Robots for Teaching Kids about Nutrition [ROMAN '14]

As part of an NSF Expedition in Computing, we completed a multi-site study in which we sought to understand how children engage with a social robot in an educational context. The study was a long-term interaction in which 26 children (aged 5-8) interacted one-on-one with a social robot in six different sessions over the course of three weeks. Each interaction involved the child learning about nutrition and how to make healthy food choices. We found that children had a positive reaction to the robot, which was maintained over the course of the multi-session interaction. Additionally, we found a statistically significant increase in engagement over time, as measured by the types of verbal responses used by the children.

Reidentification In Public Domains [PST '12]

We scraped public data from two social networking sites (Facebook and Twitter) to explore how we could reidentify individuals across sites. We looked at two different strategies, using traditional reidentification attributes versus using friendship information. We found that having access to a user's friendship information greatly increases the likelihood of reidentification across the two sites. We also demonstrated the possibility of reidentification of individuals in an anonymized dataset released by the Census Bureau using publicly available data. We concluded that targeted reidentification is fairly straightforward, given the large amount of public data available.