Mixed-Initiative Techniques for Assisted Visualization

Abstract

The advent of computers and their subsequent use in scientific applications, has led to thegeneration of a large amount of information. This information usually has two characteristics, it is large and it is multi-dimensional. The need for analysis and exploration of this data has led to investigations of methods to visualize the data. The generation of effective visualizations is a hard problem, experiments on the use of the low level human-visual system have led to a set of guidelines which can produce perceptually salient visualizations. However the application of these guidelines to generate good mappings between data attributes and visual features, requires a human to be an expert on visualization.

We have developed a semi-automated visualization assistant (ViA), that uses the perceptual guidelines and dataset properties to search the space of mappings to generate effective visualizations. However ViA's knowledge about the perceptual guidelines needs to be augmented with the user's preferences, knowledge about the dataset and the ultimate purpose of the visualization, to generate effectual visualizations.

Mixed initiative interaction is an area of research that aims to provide efficient communication between agents collaboratively solving a problem, allowing each agent to contribute what it's best at towards the solution of the problem. We have developed a distributed interaction framework for ViA that supports mixed-initiative interaction. The framework incorporates decision theory and models user preferences to provide mixed-initiative interaction in order to rapidly generate effective visualizations.