Approximately two-thirds of the U.S. adult population is overweight or obese. A main contributing factor to this problem is sedentary behavior. In this project, the research team investigated how to provide AI-based personalized digital environments to better motivate individuals for physical activities. Social comparison, a common psychological process, is known to motivate people to engage in physical activities. It has been widely used in social fitness apps through features such as “leaderboards.” However, contemporary social psychology theories indicate that these features only match the needs of a portion of the population, who are motivated by comparing themselves with people who do better than them. This project is among the first to investigate personalized social comparison environments automatically based on individual preferences. In particular, we investigated 1) how to detect which type of social comparison better motivates an individual and 2) how to automatically adapt digital environments to a group of people so they can be more active collectively.  

This project resulted in several key contributions to the primary disciplines. First, the research team iteratively developed several novel applications, two web-based apps and one multiplayer game for health called “Step Heroes”, that personalized users’ social comparison environment to motivate physical activities. Each system advanced the science of user modeling, machine learning, and the design science of human-centered AI applications. Furthermore, these applications expanded the reach of AI personalization from individual to group users. As a result, we developed new machine learning techniques (Shapley Bandits) with built-in considerations of ethical principles such as fairness. Our techniques achieved better user retention and motivation across the participants in our studies.

Second, we conducted three main rounds of user studies and collected empirical data on how over a hundred users interacted with our systems. In each user study, we recruited participants from the Drexel network, and they used various versions of our system for 21 days. The results of our analysis led to insights into how to predict individuals’ social comparison preferences and how to design AI-based personalizations. Importantly, critical reflections on our analysis revealed a phenomenon underlying AI personalization, which we called the “paradox of personalization,” where user modeling and user adaptation formed a distorting feedback loop. 

Third, in addition to contributions to human-computer interaction (HCI) and AI, empirical results from our project have led to contributions to social psychology and the promotion of physical activity. Our journal article in JMIR Human Factor compared data from multiple user studies from this project and confirmed that day-to-day differences in social comparison preferences are associated with changes in physical activity motivation and behavior. 

We published 16 peer-reviewed scientific papers in various high-impact venues, including the ACM Computer-Human Interaction (CHI) Conference, ACM Computer Supported Cooperative Work and Social Computing (CSCW), ACM Intelligent User Interface (IUI), Journal of Medical Internet Research (JMIR) Human Factors, Foundations of Digital Games (FDG), ACM CHI Play, and IEEE Conference of Games. One of the papers received a Best Paper Award at the 2021 Conference of the Foundations of Digital Games (FDG’21). The PIs also gave invited talks and keynotes about the project at various universities and conferences.

Last Modified: 12/26/2023
Modified by: Jichen Zhu