Knowledge Generation in Visual Analytics

可视化分析（VA）结合了人类和机器的能力，以便从数据中生成知识。系统能够处理大量数据，而人类则通过利用知识来推理问题。人类通过与系统的交互引导分析过程来参与VA过程。大多数当前的方法有两个主要的缺点：1）一方面，分析人员不能在分析过程中外部化他们的专业知识; 2）另一方面，他们不理解系统中发生的过程。该项目旨在将人类和机器更紧密地联系在一起，以增强VA，从而实现更有效和更高效的数据分析。这将通过将数据和分析来源与自动化和可视化方法联系起来来实现。这两种类型的来源信息将用于支持和自动化人类知识生成过程，根据用户在当前分析阶段的需求自适应。这项研究计划的一个主要部分是弥合人类和机器学习（ML）之间的差距，以便使复杂的模型配置和交互更容易访问和使用。此外，我们的目标是减少在分析过程中由认知偏差引起的人为错误。这台机器可以作为一个公正的对手。本研究提案的独特之处在于对VA的整个知识生成过程进行了全面的透视，其目标是通过沿着整个VA管道开发方法来提高最先进的水平。研究的方法将被应用到几个真实的世界的数据集，域，任务，和用户在分析领域的飞行轨迹，政治辩论，和子空间聚类的高维数据。这项研究的好处将通过用户研究来评估，这些研究说明了新方法的好处，这些新方法可以增强数据分析过程，使其更容易获得，更有效，更高效，更透明和更可靠。

Visual Analytics (VA) combines human and machine capabilities in order to generate knowledge from data. Systems are capable of processing large amounts of data while humans reason about their problems by leveraging their knowledge. Humans participate in the VA process by steering the analysis process through interactions with the system. Most current approaches have two major drawbacks: 1) on the one hand, analysts cannot externalize their expert knowledge during the analysis process; 2) on the other hand, they do not understand the processes happening in the system. The project aims at bringing human and machine closer together in order to enhance VA for a more effective and efficient data analysis. This will be achieved by bridging data and analytic provenance with automated and visualization methods. The two types of provenance information will be used to support and automate human knowledge generation processes adaptively according to the users´ needs in their current analysis stage. A major part of this research proposal is to bridge the gap between human and machine learning (ML) in order to make complex model configuration and interaction more accessible and usable. Further, we aim to mitigate human errors during analysis processes caused by cognitive biases. The machine can be used as an unbiased counter party. The unique feature of this research proposal is a holistic perspective onto the entire knowledge generation process in VA with the goal of enhancing the state of the art by developing methods along the entire VA-pipeline. The investigated methods will be applied to several real world datasets, domains, tasks, and users in the analysis areas of flight trajectories, political debates, and subspace clustering in high-dimensional data. The benefits of this research will be evaluated through user studies illustrating the benefits of the novel methods, which enhance the data analysis process to be more accessible, effective, efficient, transparent, and reliable.