tmfast fits topic models fast

TL;DR

tmfast is an R package that enables rapid fitting of topic models by leveraging a novel algorithm based on partial PCA and varimax rotation, improving efficiency while maintaining model quality.

Contribution

The paper introduces a new fast algorithm for topic modeling implemented in the tmfast R package, with mathematical background and empirical comparisons.

Findings

tmfast significantly reduces computation time

Models fitted with tmfast are comparable in quality to standard methods

Demonstrated effectiveness on simulated and historical text data

Abstract

tmfast is an R package for fitting topic models using a fast algorithm based on partial PCA and the varimax rotation. After providing mathematical background to the method, we present two examples, using a simulated corpus and aggregated works of a selection of authors from the long nineteenth century, and compare the quality of the fitted models to a standard topic modeling package.