English

Prediction Focused Topic Models via Feature Selection

Machine Learning 2020-03-04 v2 Computation and Language Machine Learning

Abstract

Supervised topic models are often sought to balance prediction quality and interpretability. However, when models are (inevitably) misspecified, standard approaches rarely deliver on both. We introduce a novel approach, the prediction-focused topic model, that uses the supervisory signal to retain only vocabulary terms that improve, or at least do not hinder, prediction performance. By removing terms with irrelevant signal, the topic model is able to learn task-relevant, coherent topics. We demonstrate on several data sets that compared to existing approaches, prediction-focused topic models learn much more coherent topics while maintaining competitive predictions.

Keywords

Cite

@article{arxiv.1910.05495,
  title  = {Prediction Focused Topic Models via Feature Selection},
  author = {Jason Ren and Russell Kunes and Finale Doshi-Velez},
  journal= {arXiv preprint arXiv:1910.05495},
  year   = {2020}
}

Comments

AISTATS 2020. arXiv admin note: substantial text overlap with arXiv:1911.08551

R2 v1 2026-06-23T11:41:46.329Z