A framework for the formalization of interactive visual analytics

点击翻译按钮获取中文摘要

The process of visualization is about transforming data into pictures. The computer science study of visualization is about the development of the theory and practise of transforming data into pictures from which humans can draw inferences (we use the word "picture" to include all manner of visual media, from dots on a page to 3D video). Because the variety and breadth of the world's data is enormous, a visualization process must necessarily filter, compress, or otherwise reduce the scope and complexity of data in order to transform it into sensible pictures. So one foundational challenge of visualization is to ensure that data to picture transformations preserve those properties that best help humans draw "appropriate" inferences about those data. For example, a bar chart that represents the number of hockey players by country of origin should make it easy to see which country produces the most hockey players. In fact, one measure of a good picture is that it leads most if not all humans to draw the same conclusions, and that it avoids introducing visual anomalies (e.g., like the ambiguity of a Necker cube). A major challenge is about how to build transformations that preserve important properties of the base data (e.g., numbers in a spreadsheet) when turning them into a picture (e.g., a histogram). Since visualization can't merely present every base data point in a picture, some aggregation of data is required. For example, if our hockey player country of origin transformed the ages of players into averages, one could likely "see" which players were on average younger, but not find out which was youngest or oldest. The most important visualization research challenges are about how to reduce data volumes to visually manageable forms, how to determine what aggregate properties are important to preserve in that transformation, and then how to evaluate alternative transformations by confirming the most typical inferences by humans. The recent almost ubiquitous use of modern touch screen technologies exacerbates the challenge: in addition to preserving data properties, reducing ambiguity, and confirming preferred pictures for efficient human inference, the question of visual manipulation begs the challenge of appropriate repertoires of picture manipulation, and how well they can help reveal data properties within pictures. Overall, the challenge of visualization research is to consider property preservation, avoidance of ambiguity, confirmation of best inference support, and identification of appropriate repertoires of picture actions to improve human understanding of data. All advances help provide higher value exploitation of all forms of scientific and business data.