Designing Social Choice Mechanisms Using Machine Learning

Social choice studies ordinal preference and information aggregation with applications in high-stakes political elections as well as low-stakes movie rating websites. Recently, computational aspects of classical social choice mechanisms have been extensively investigated, yet not much has been done in designing new mechanisms with the help of computational techniques. In this paper, we outline a workflow to formalize a principled approach towards designing novel social choice mechanisms using machine learning. In the workflow, we clearly separate the following two goals of social choice (1) reaching a compromise among agents' subjective preferences, and (2) revealing the ground truth. For each of the two goals, we discuss criteria for evaluation, main challenges, possible solutions, and future directions.