Preference Elicitation For General Random Utility Models

TL;DR

This paper introduces two Bayesian-based preference elicitation schemes for General Random Utility Models (GRUMs), improving estimation precision through a Monte-Carlo-EM algorithm and theoretical likelihood analysis.

Contribution

It presents novel preference elicitation methods for GRUMs and a Monte-Carlo-EM algorithm for MAP inference, with proven likelihood unimodality.

Findings

Elicitation schemes increase estimation precision

Proposed algorithms effectively infer preferences

Likelihood functions are unimodal for certain GRUMs

Abstract

This paper discusses {General Random Utility Models (GRUMs)}. These are a class of parametric models that generate partial ranks over alternatives given attributes of agents and alternatives. We propose two preference elicitation scheme for GRUMs developed from principles in Bayesian experimental design, one for social choice and the other for personalized choice. We couple this with a general Monte-Carlo-Expectation-Maximization (MC-EM) based algorithm for MAP inference under GRUMs. We also prove uni-modality of the likelihood functions for a class of GRUMs. We examine the performance of various criteria by experimental studies, which show that the proposed elicitation scheme increases the precision of estimation.