Skip to content


Repository files navigation

Topic Model Evaluation


This repository contains code for replicating the topic model evaluation experiments in Scott & Baldridge 2013, "A recursive estimate for the predictive likelihood in a topic model."

The code in this repository contains implementations for the evaluation methods discussed in the paper, data management code for the corpora we used, and glue code to train topic models using the Mallet toolkit. There is code in R that is used for the smaller simulated experiments (section 4.1) and Scala code for the larger simulations (section 4.2) and experiments with actual text corpora (seciton 4.3). See the wiki page for instructions for using the code to reproduce the tables and and figures in the paper.

You can see the topics computed for the six corpora. If you don't know what a topic model is and are curious, check out the resources on David Blei's topic modeling page and the Mallet page on topic modeling.

If you have any problems running the code, please create an issue.

It is possible that updates will be made to this repository (e.g. if there are any fixes or added documentation). To get the exact version that was used to complete the experiments done in the paper, check out the tag v1.0, or you can download that tagged version directly.


Code for topic model evaluation experiments for Scott & Baldridge 2013, AISTATS







No packages published