Guidance in Visual Analytics through Mixed-Initiative and Implicit Interaction

可视化分析是一门结合数据处理和信息可视化来支持人们理解数据的科学。它的强大之处在于让计算机做与其能力相适应的工作，比如计数和组织数据，而让人类分析师做推理、收集证据和决策。长期以来，这种协作的计算方面的重点一直放在数据挖掘、布局和显示大量信息上。指导分析过程主要是人类分析师的专属领域。我的研究将探讨软件如何在探索数据集时建议人类分析师采取的步骤。很多人际交往都是通过含蓄的方式进行的——肢体语言、语调、面部表情和用词。我们的显性交流是由这些隐性线索支持的。隐性线索提供了很多机会来提高我们使用计算机系统的工作效率。例如，通过监测用户注视图表中一系列点的目光，分析系统可以在没有明确询问和用户明确陈述兴趣的情况下，建议其他未探索的项目，这些项目可能符合用户的兴趣。这项研究将探索笔、触摸、虚拟现实中的手势和眼睛注视的隐含线索。在分析数据方面，计算机的作用发挥到什么程度，这是一个权衡。自动突出显示或隐藏数据会带来算法偏差的风险。但在没有指导的情况下给分析师增加负担同样是不合适的，因为这项工作更加困难，而且重要数据可能会被遗漏。我将研究这种权衡：控制应该在哪里，什么样的指导是可能的和有用的，什么时候应该提供指导？文本分析将是调查的主要测试平台。每天都有数十亿页不同语言的文本产生：电子邮件、新闻、博客、评论、论坛、学术文章和商业报告。这项研究将产生一系列文本交互工具，这些工具结合了隐含的交互、复杂的语言分析和可视化来揭示隐藏的信息，如主题、趋势和情绪。我设想引入新的方法来浏览和使用在线文本集合，通过丰富和集中分析和阅读的行为来减轻信息过载的负担。这项研究将引入新的可视化分析实践，实现分析师和算法的更紧密耦合。12名受训者将接受数据科学技能方面的专门培训。结果有望提高加拿大人从从商业智能到健康信息学等领域的经济和社会重要大规模数据集中获得见解的能力。该研究结合了安大略省商业开放议程的交互式数字媒体和人工智能主题，直接解决了政府的优先事项。

Visual analytics is the science of combining data processing and information visualization to support people in understanding data. Its strength comes from having computers do work appropriate to their capabilities, such as counting and organizing data, while human analysts do the work of reasoning, evidence gathering, and decision-making. The focus of the computational aspect of this collaboration has long been on data mining, layout, and displaying large amounts of information. Guiding the analytic process has primarily been solely the domain of the human analyst. My research will investigate ways for software to advise human analysts on steps to take when exploring datasets. Much interpersonal communication occurs implicit means - body language, intonation, facial expression, and word choice. Our explicit communication is supported by these implicit cues. Implicit cues offer a lot of opportunity to improve the effectiveness of our work with computer systems. For example, by monitoring the gaze of a user looking at a sequence of points in a chart, an analytic system could suggest additional unexplored items which may fit the user's interest, without being explicitly asked, and without the user explicitly stating their interest. This research will explore implicit cues from pen, touch, gestures in virtual reality, and eye gaze.  There is a trade-off around how far to take the role of computers in analyzing data. Automatic highlighting or hiding of data introduces risk of algorithmic bias. But burdening analysts without guidance is equally inappropriate, as the work is more difficult, and important data may be missed. I will investigate this trade-off: where should the control lie, what sorts of guidance is possible and helpful, and when should guidance be provided? Text analytics will be the main testbed for the investigation. Billions of pages of text, in many languages, are produced daily: emails, news, blogs, reviews, discussion forums, academic articles, and business reports. This research will produce a collection of interactive tools for text which couple implicit interaction, sophisticated linguistic analysis, and visualization to reveal hidden information, such as topics, trends, and emotions. I envision introducing new ways to navigate and use text collections online, lessening the burden of information overload by enriching and focusing the acts of analysis and reading. This research will introduce new visual analytic practices, achieving a closer coupling of analysts and algorithms. 12 trainees will receive specialized training in highly sought-after data science skills. The outcomes promise to improve Canadians' ability to gain insights from economically and socially important large-scale datasets in domains from business intelligence to health informatics. Blending the interactive digital media and artificial intelligence themes of Ontario's Open for Business Agenda, this research directly addresses government priorities.