Title: A simple generative model for rank ordered data with ties
Authors: Daniel Henderson - Newcastle University (United Kingdom) [presenting]
Abstract: A model for ranked ordered data which naturally accommodates ties is proposed. The model is inspired by the exponential latent variable formulation of the Plackett-Luce model and can be seen as its discrete counterpart. Specifically, the generative model assumes that the data arise from independent geometric latent variables. The latent variables can be integrated out of the model analytically, resulting in a simple likelihood function that facilitates straightforward inference. With a focus on Bayesian inference, a simple Gibbs sampling algorithm is presented. Several extensions of the basic model are considered.