Home Page

Papers

Submissions

News

Editorial Board

Special Issues

Open Source Software

Proceedings (PMLR)

Data (DMLR)

Transactions (TMLR)

Search

Statistics

Login

Frequently Asked Questions

Contact Us



RSS Feed

Principled Selection of Hyperparameters in the Latent Dirichlet Allocation Model

Clint P. George, Hani Doss; 18(162):1−38, 2018.

Abstract

Latent Dirichlet Allocation (LDA) is a well known topic model that is often used to make inference regarding the properties of collections of text documents. LDA is a hierarchical Bayesian model, and involves a prior distribution on a set of latent topic variables. The prior is indexed by certain hyperparameters, and even though these have a large impact on inference, they are usually chosen either in an ad-hoc manner, or by applying an algorithm whose theoretical basis has not been firmly established. We present a method, based on a combination of Markov chain Monte Carlo and importance sampling, for estimating the maximum likelihood estimate of the hyperparameters. The method may be viewed as a computational scheme for implementation of an empirical Bayes analysis. It comes with theoretical guarantees, and a key feature of our approach is that we provide theoretically-valid error margins for our estimates. Experiments on both synthetic and real data show good performance of our methodology.

[abs][pdf][bib]       
© JMLR 2018. (edit, beta)

Mastodon