TRELLIS: VISUAL TOPIC AGGREGATION

Abstract:

With increasing availability of large volumes of text data, researchers from many fields have more documents than they can easily analyze by hand. Topic models are powerful computational tools that can be of great benefit to these researchers, but technical requirements can make these models inaccessible to a broad audience. With the goal of empowering users to examine their text data and refine models of their corpora based on domain exper- tise, Trellis is a visual tool for topic model curation and dataset exploration. By allowing researchers to visually interact with their datasets and aggregate topics within the model, Trellis enables an iterative process of adjusting a working hierarchical topic model and examining the corresponding dataset based on that working model.