DOCUMENT EXPLORATION WITH TOPIC MODELING : DESIGNING INTERACTIVE VISUALIZATIONS TO SUPPORT EFFECTIVE ANALYSIS WORKFLOWS
Uwakmfon
Document Exploration with Topic Modeling : Designing Interactive Visualizations to Support Effective Analysis Workflows
We present a series of published and ongoing projects, on developing interactive visualizations for document exploration using statistical topic models. Taking a user-centric approach, we engage analysts working with real-world datasets, identify bottlenecks in their analysis workflow, and respond by developing appropriate visual presentations, interactions, and algorithms to address the users’ needs. We first review visual analysis tools produced by the authors of this paper. While these tools have been previously published in various venues, here we summarize their collective contributions, report on user feedbacks received since the deployment of the tools, and identity potential areas of improvement. We then present various visualization components currently under development, and report on informal feedbacks received from early prototypes. Effective visual designs often need to be paired with appropriate algorithmic development. We highlight potential areas of collaboration between machine learning and visualization researchers, to address the present barriers to analysis. Finally, we outline how a completed system may enable evaluations of topic modeling based on task performance.
