A Double Parametric Bootstrap Test for Topic Models

TL;DR

This paper introduces a double parametric bootstrap test to evaluate the fit of NMF-based topic models, addressing likelihood assumption violations in real document corpora.

Contribution

It proposes a novel bootstrap testing method leveraging KL divergence and Poisson likelihood to assess NMF topic model reliability.

Findings

Test accurately identifies well-fitting models in simulations

Effective in real-world document datasets

Addresses likelihood assumption violations

Abstract

Non-negative matrix factorization (NMF) is a technique for finding latent representations of data. The method has been applied to corpora to construct topic models. However, NMF has likelihood assumptions which are often violated by real document corpora. We present a double parametric bootstrap test for evaluating the fit of an NMF-based topic model based on the duality of the KL divergence and Poisson maximum likelihood estimation. The test correctly identifies whether a topic model based on an NMF approach yields reliable results in simulated and real data.