Latent Dirichlet Allocation

When we write an article, we develop it according to a theme or topic and we use certain words from that topic. We may have sub-topics and use the words for the sub-topics too. When we classify articles into topic piles, we recognize specific words and then tag them so that we can place them into topics. An article may have one topic and other sub-topics, so it is possible to tag an article to multiple topics. Latent Dirichlet Allocation (LDA) is designed to discover abstract topics in a document. This makes LDA a powerful model that can tag an article with multiple topics.

LDA is the core technique in NLP and is worth investigating thoroughly. It has enabled many commercial products. The knowledge of LDA, such as its architecture, its use of the generative modeling process, and Dirichlet distribution, are transferable to other models. For these reasons, this book has dedicated four chapters to LDA: Chapter 9, Understanding Discrete Distributions which...