A Simulation Study of Topic Model Performance under Varying Corpus Conditions
Date:
Poster presentation (Poster #1, Session 2) at the BaYSM - Bayesian Young Statisticians Meeting on two Monte Carlo simulation studies examining LDA and STM topic model performance under varying corpus conditions. Study I investigated LDA sensitivity in small corpora; Study II compared LDA and STM under temporally evolving topic prevalence, revealing a counterintuitive crossover in model performance favourable to LDA for longer documents.
Joint work with John Quigley and Matthew Revie (University of Strathclyde).