This Faculty Early Career Development Program (CAREER) award supports research on novel, consumer behavior aware design of online marketplaces through advances in online machine learning techniques. This grant will benefit U. S. consumers and service providers and contribute to an increased competitiveness of the e-commerce sector by addressing critical challenges faced in revenue management and operations of modern marketplaces. With the ubiquity of online customer review platforms, price aggregators, discussion forums, and connected communities, the modern consumer displays increasing tendencies to form socially influenced opinions, imitate other users, learn and anticipate prices, and time their purchases. Market mechanisms that fail to account for such consumer behavior can result in significant losses in terms of revenue, inventory, and consumer welfare. This project will leverage the opportunity presented by the abundance of consumer activity data to learn and account for consumer behavior when designing market pricing and allocation mechanisms. The research conducted will benefit both consumers and sellers through a more efficient matching of demand and supply. The educational and outreach initiatives of this program will enable inclusive participation and engagement of students at all levels, and promote interdisciplinary, hands-on education in data science and decision making.
The research plan of this project addresses some critical gaps in the current online machine learning techniques, which are typically designed assuming independent and identically distributed or exogenously generated data; and therefore, are ill suited to handle the endogeneity that may result from frequent social interactions and strategic consumer. The research plan investigates new learning techniques for dynamic pricing, recommendation, and allocation under three broad classes of commonly occurring consumer behavior: (1) social influence and imitation, (2) forward-looking and planning behavior, and (3) consumer learning behavior. For each of these categories, it will develop new models and decision-making frameworks that can capture natural and uncertain consumer behavior while being conducive to learning. New methods for exploration-exploitation, dynamic learning, and online optimization will be developed, which will build upon and advance tools from multi-armed bandits, online convex optimization, dynamic programming, and reinforcement learning. These new solution techniques will be accompanied with rigorous theoretical analysis as well as performance evaluation in real markets through collaboration with industry partners.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
