Software

An R package for estimation of the Structural Topic Model, available on CRAN and GitHub with a project website and vignette.

An explainable inference-time-compute framework for local LLMs. STATe-of-Thoughts searches over high-level reasoning patterns using structured action templates, tree search, and evaluation of intermediate reasoning steps.

An R package implementing design-based supervised learning, a framework for using predicted variables in downstream statistical analyses while obtaining valid estimates and standard errors.

An R package for estimating context-specific word and short-document embeddings and evaluating hypotheses about covariate effects on embeddings.

A Python implementation of the SUpervised Nonnegative matrix factorization topic model for topic discovery and effect estimation, developed for the AutoPersuade project.

An R package for validating topic coherence and topic labels using crowd-sourcing workflows that can be posted to and retrieved from Amazon Mechanical Turk.

An R package for memory-efficient optimization with an L1 penalty, wrapping libLBFGS for L-BFGS and OWL-QN.

A D3 visualization for interactively exploring relationships between topics and covariates estimated from stm.

An interactive visualization of topic correlations and hierarchy in a Structural Topic Model.

A graphical interface via RStudio Shiny for creating Structural Topic Models.

An R package for adjustment with text-based confounding, associated with the text matching approach for high-dimensional text data.