<ul><li>Topic models often struggle to capture minority topics like mental health themes in online comments.</li><li>Proposed solution involves Constrained Non-negative Matrix Factorization (NMF) to incorporate seed word lists for minority content without detailed pre-specification.</li><li>The Constrained NMF, applied with Karush-Kuhn-Tucker conditions, outperforms baselines on topic purity and mutual information on synthetic data.</li><li>In a case study on YouTube vlog comments, the model successfully identifies and highlights minority mental health content in viewer discussions.</li></ul>

Constrained Non-negative Matrix Factorization for Guided Topic Modeling of Minority Topics

Discover more