Topic modeling via scatter/gather clustering

Latent variable models such as Latent Dirichlet Allocation provide rich tools for analyzing large document corpora. They can uncover a wide range of hidden information such as topics in text, communities in social networks, and patterns in images. Scatter/Gather is a clustering technique that allows users to interactively combine and split groups. When joined with latent variable models, Scatter/Gather organizes topics into themes, enables topic browsing, and improves processing time for large numbers of topics.