CAREER: Analyzing Interactions in Visual Analytics for User and Data Modeling

Tufts University, Medford MA

Investigators

Abstract

Visual analytics systems combine human analysis with computational techniques. When the two are fully integrated these next generation human-in-the-loop systems can be tremendously powerful, but current visual analytics systems suffer from a significant communication gap between the human and the computer which prevents them from reaching their full potential. Specifically, there currently exist no generalized techniques to enable the computer to easily and intuitively understand the user's background knowledge, analysis process, or intent during an analysis task. Lacking this, the computer has limited means to identify the user's needs and provide timely and appropriate computational support. The PI's goal in this project is to develop computational techniques to quantify and extract the user's high-level knowledge by analyzing his/her interactions with a visual interface. Toward this end, the research agenda consists of two complementary components: data modeling and user modeling. In data modeling, the user's interactions are used to learn the parameters of an algorithm, computational process, or representation of the data that best reflect the user's domain knowledge about the data. In user modeling, the same interactions are used to infer aspects of the user's reasoning process and cognitive style. Together, the results of these two modeling processes will give the computer the means to better understand the user's analysis process and will enable it to better support the user in performing his/her task. The results of this work will have both immediate and long term impact on the research and application of visual analytics. In the short term new techniques for data and user modeling will advance existing systems and practices in interactive data analysis, while in the long term project outcomes will help establish the foundations of a human+computer approach to visual analytics that more effectively supports the user's analysis process, which in turn will impact the development of real-time, mixed-initiative visual analytics systems for addressing the challenges of big data. In addition, through an integrated research and education agenda, students will be trained with the appropriate balance of expertise in human reasoning and computational science, preparing them to conduct independent interdisciplinary research and lead future efforts in visual analytics.