Topic Models: Understanding Latent Structures

This lecture delves into the concept of topic models, focusing on understanding latent structures within data. Starting with a recap on clustering and density estimation, it progresses to Gaussian density estimation and Gaussian mixture models. The lecture then explores the latent variable interpretation of GMMs, learning algorithms, and limitations. It concludes with an in-depth look at Latent Dirichlet Allocation (LDA) as a form of dimensionality reduction and its generative process. The lecture also covers variational inference and the evaluation of LDA models using perplexity. Additionally, it discusses the application of LDA in Digital Humanities and provides resources for further exploration.