"Big picture": mixed-initiative visual analytics of big data

Abstract

Information graphics have been used for thousands of years to help illustrate ideas and communicate information. However, it requires skills and time to hand craft high-quality, customized information graphics for specific situations (e.g., data characteristics and user tasks). The problem becomes more acute when we must deal with big data. To address this problem, we are researching and developing mixed-initiative visual analytic systems that leverage both the intelligence of humans and machines to aid users in deriving insights from massive data. On the one hand, such a system automatically guides users to perform their data analytic tasks by recommending suitable visualization and discovery paths in context. On the other hand, users interactively explore, verify, and improve visual analytic results, which in turn helps the system to learn from users' behavior and improve its quality over time. In this talk, I will present key technologies that we have developed in building mixed-initiative visual analytic systems, including feature-based visualization recommendation and optimization-based approaches to dynamic data transformation for more effective visualization. I will also use concrete applications to demonstrate the use and value of mixed-initiative visual analytic systems, and discuss existing challenges and future directions in this area.