Building and Exploring Dynamic Topic Models on the Web

Topic modeling is a machine learning technique that identifies latent topics in a text corpus. There are several existing tools that allow end-users to create and explore topic models using graphical user interfaces. In this paper, we present a visual analytics system for dynamic topic models that goes beyond the existing breed of tools. First, it decouples the Web-based user interface from the underlying data sets, enabling exploration of arbitrary text data sets in the Web browser. Second, it allows users to explore dynamic topic models, while existing tools are often limited to static topic models. Finally, it comes with a tool server in the backend that allows the design and execution of scientific workflows to build topic models from any data source. The system is demonstrated by building and exploring a dynamic topic model of CIKM proceedings published since 2001.