Value-Based Voting Semantics for Multi-Agent Qualitative Preference Reasoning

Qualitative preference reasoning has emerged as an effective framework for representing and reasoning with user preferences, as it allows users to express preferences over the properties of choices rather than the choices themselves. This approach simplifies preference elicitation and enables efficient reasoning over a large set of alternatives by focusing on a relatively small set of their shared properties. While the semantics of qualitative preference reasoning are well established for single-agent settings, limited progress has been made in addressing scenarios involving multiple agents with potentially conflicting preferences. This thesis investigates the use of voting-based semantics to aggregate and reason with qualitative preferences in multi-agent contexts. The distinguishing feature of our approach is that each vote considers not only whether it supports one choice over another, but also the weight of the agent casting the vote and the conviction/strength associated with that preference. We propose ten distinct semantics grounded in Pareto, majority, maximal, and logical principles, and we conduct experiments to evaluate their effectiveness in capturing the dynamics of group decision-making.

Committee: Samik Basu (major professor), Pavan Aduri, and Myra Cohen