Online Program Home
  My Program

Abstract Details

Activity Number: 172 - Machine Learning and Algorithms
Type: Contributed
Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistical Computing
Abstract #323160 View Presentation
Title: A Moment-Based Estimation Procedure for Deeply-Nested Hierarchical Models
Author(s): Ningshan Zhang* and Patrick Perry and Kyle Schmaus
Companies: New York University and New York University and Stitch Fix
Keywords: Recommender Systems ; Hierarchical Model ; Generalized Linear Mixed Model ; Computational Efficiency
Abstract:

In this talk, we consider the problem of using a database of book reviews to inform user-targeted recommendations. In our dataset, books are categorized into genres and sub-genres. We use a hierarchical model, which is able to exploit this nested structure to pool information across similar items at many levels within the genre hierarchy simultaneously. Our main challenge is that fitting our model at scale using off-the-shelf maximum likelihood procedures is prohibitive due to the large data sizes involved. To get around this, we extend a moment-based fitting procedure proposed by Perry in 2016 for fitting two-level hierarchical models, which is an order of magnetite faster than maximum likelihood. Our procedure can be used in other contexts for fitting deeply-nested hierarchical generalized linear mixed models efficiently.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2017 program

 
 
Copyright © American Statistical Association