CAREER: Visual Analytics by Demonstration for Interactive Data Analysis

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

In today's data-driven era, more people analyze data as part of their daily lives. Visual analytic technologies help people in a variety of these domains including business, health, education, national security, and many more. These visual analytic techniques are effective since they combine powerful machine learning with interactive data visualization. Currently, the way that people use such systems is through user interfaces with control panels to adjust parameters of the analytic models and visualization properties directly. For example, controls may ask users to select which analytic model to use, which parameter to adjust, and by how much. However, when people do not have the necessary expertise or training in data science or information visualization, they may not be able to properly use these tools and miss out on important insights. Instead, this project will explore, design, develop, and evaluate techniques that allow people to demonstrate their analytic goals, tasks, and operations. Developing this "visual analytics by demonstration" method of user interaction has the potential to impact numerous data-driven domains, and society more broadly. This project will also provide educational experience and research training for graduate and undergraduate students to guide them towards careers in computing. The proposed research will create Visual Analytics by Demonstration prototypes, generalizable toolkits, and demonstration primitives to foster exploration and discovery in visual analytics. Instead of control panels that require users to directly parameterize analytic models and visualizations, people provide demonstrations, from which the system selects the appropriate visual representation, analytic model, and parameters. To realize the benefits of such technology, many research challenges exist and must be addressed. For instance, what are the basic demonstration primitives that people use to communicate their intent to a system? How can systems interpret these demonstrations and perform the correct analytic and visualization operations? Finally, how can these systems guard against potential user bias in exploring data using by-demonstration? Project objectives include the design, implementation, and evaluation of by-demonstration visual analytic prototypes. The research will perform formative studies to develop demonstration primitives categorized by user tasks. The project will design and develop applications for specific data domains as well as general, open-sourced toolkits for other researchers to use and extend. User studies will identify if and how specific tasks and operations can be performed by demonstrations, and how these compare in performance to current control panel interfaces. The proposed work will also create instructional material to integrate visual analytics by demonstration into courses that teach visual analytics, or data science more generally. The project website (http://va.gatech.edu/projects/visual-analytics-by-demonstration/) will include project information, links to resulting publications, videos, and open-source software produced. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.