B1646
Title: Mixtures of generalized Plackett-Luce models
Authors: Daniel Henderson - Newcastle University (United Kingdom) [presenting]
Abstract: The problem of clustering rank-ordered data with ties is addressed via a mixture of generalized Plackett-Luce models. The generalized Plackett-Luce (GPL) model is for ordered partitions based on order statistics of latent geometric random variables. Such a view facilitates straightforward inference algorithms for maximum likelihood and Bayesian estimation. These algorithms can be extended naturally to consider finite mixtures of GPL models, leading to a framework for model-based clustering of rank-ordered data with ties. Several illustrative examples are provided.