CHS: Small: Collaborative Research: Representing and Learning Visualization Design Knowledge

该项目提供了新的方法和软件工具，用于创建数据驱动的可视化，以提高可视化分析和数据通信的清晰度和有效性。许多可视化设计准则，如“避免高度饱和的颜色”或“从0开始条形图中的条形”，都源于人们如何阅读各种类型的可视化的实证研究。然而，这些准则往往是在书籍或文章中非正式地陈述。在设计可视化时，作者可能不得不做出决定，将一个设计准则优先于另一个，但这些原则的非正式性质并没有为如何做到这一点提供足够的指导。即使可视化研究人员和系统设计人员在更正式的“知识库”中表示设计指南，创作系统可以使用这些知识库来指导可视化作者获得更有效的图形，这些指南也是基于一个人仔细总结经验结果，这是一个容易出错的过程。本项目通过创建识别、聚合、编辑、测试和搜索可视化设计知识的新方法，解决了制定和应用可视化设计知识的这些挑战。这项研究还将解决现有可视化设计知识的差距，应用新的方法来制定和评估设计指南，以创建有效的“多视图”可视化（如分析仪表板或顺序演示），可视化非常大的数据集，并直观地表达数据中的不确定性或错误。我们将创建包含这些类型的可视化指导方针的知识库，以及一个创作工具，以帮助作者在设计可视化（如仪表板）时管理单个视图和多个视图之间的竞争设计考虑因素。本研究开发的所有实验结果、知识库和创作工具都将免费公开。为了实现这些目标，本项目开发了一套方法，用于从可视化感知和解释的实证研究中识别和评估可视化设计指南。为此，该团队将开发方法来重新表达图形感知和认知相关实验文献的现有结果作为约束，并创建新的方法和工具，直接从可视化专家（如熟练的设计师或研究人员）那里获得设计指南。该项目还将产生用于生成可视化和收集特定任务的可视化判断的自动化方法，以便为给定的一组设计约束学习适当的优先级权重。通过开发用于捕获经验结果的表示和模型，这些结果可以解释人类受试者实验结果中固有的不确定性，该项目旨在综合和澄清有关可视化设计的现有经验知识。该研究还将通过贡献基本方法来推进可视化设计知识的最新发展，这些方法用于（1）识别和学习大型数据集可视化，多视图可视化（如仪表板）和不确定性可视化的指导方针，以及（2）探索用于浏览，编辑，该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project contributes new methods and software tools for creating data-driven visualizations that improve the clarity and effectiveness of visual analysis and communication of data. Many visualization design guidelines, like "avoid highly saturated colors", or "start bars in a bar chart at 0", stem from empirical studies of how well people can read visualizations of various types. However, these guidelines are often stated informally in books or articles. In designing a visualization, an author may have to make decisions that prioritize one design guideline over another, yet the informal nature of such principles does not provide sufficient guidance for how to do this. Even when visualization researchers and system designers represent design guidelines in more formal "knowledge bases" that an authoring system can use to guide visualization authors towards more effective graphs, the guidelines are based on a person carefully summarizing the empirical results, an error-prone process. This project addresses these challenges to formulating and applying visualization design knowledge by creating new methods to identify, aggregate, edit, test, and search visualization design knowledge. This research will also address gaps in existing visualization design knowledge, applying novel methods to formulate and assess design guidelines for creating effective "multiple-view" visualizations (such as analysis dashboards or sequential presentations), visualizing very large datasets, and visually expressing uncertainty or error in data. We will create knowledge bases containing guidelines for these types of visualizations as well as an authoring tool to help authors manage competing design considerations between single and multiple views when designing visualizations like dashboards. All experimental results, knowledge bases, and authoring tools developed in this research will be made freely and publicly available.To meet these goals, this project develops a set of methods for identifying and evaluating visualization design guidelines from empirical research on visualization perception and interpretation. To do this, the team will develop ways to re-express existing results from relevant experimental literature on graphical perception and cognition as constraints, and create new methods and tools for directly eliciting design guidelines from visualization experts such as skilled designers or researchers. The project will also produce automated methods for generating visualizations and collecting task-specific visualization judgments in order to learn appropriate priority weights for a given set of design constraints. By developing representations and models for capturing empirical results that can account for the uncertainty that is inherent in results from human subjects experiments, the project stands to synthesize and clarify existing empirical knowledge about visualization design. The research will also advance the state of the art in visualization design knowledge by contributing fundamental methods for (1) identifying and learning guidelines for large dataset visualizations, multiple view visualizations like dashboards, and uncertainty visualizations, and (2) exploring effective interface designs for browsing, editing, and testing visualization knowledge bases.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.