A Multiparameter Mallows Model for Infinite Rankings

• Activity Number: 20
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Marina Meila*
• Companies: University of Washington
• Keywords: permutations ; partial ordering ; Mallows model ; distance based ranking model ; exponential family ; branch-and-bound

Abstract

We study the natural extension of stagewise ranking to the the case of countably many items. We introduce the infinite version of the generalized Mallows model of (Fligner and Verducci, 1986), give procedures to estimate its parameters and central permutation from data, and demonstrate that it has sufficient statistics, being thus an exponential family model with continuous and discrete parameters. The experiments demonstrate that the Infinite Mallows Model can be tractably and usefully applied to truncated rankins of very large sets of items.

Joint work with Le Bao.