Poetzsch Tristan, Germanakos Panagiotis, Huestegge Lynn
Department of Psychology, Julius-Maximilians-University Würzburg, Würzburg, Germany.
User Experience ICD, Product Engineering, Intelligent Enterprise Group, SAP SE, Walldorf, Germany.
Front Artif Intell. 2020 Mar 20;3:9. doi: 10.3389/frai.2020.00009. eCollection 2020.
Data analytics as a field is currently at a crucial point in its development, as a commoditization takes place in the context of increasing amounts of data, more user diversity, and automated analysis solutions, the latter potentially eliminating the need for expert analysts. A central hypothesis of the present paper is that data visualizations should be adapted to both the user and the context. This idea was initially addressed in Study 1, which demonstrated substantial interindividual variability among a group of experts when freely choosing an option to visualize data sets. To lay the theoretical groundwork for a systematic, taxonomic approach, a user model combining user traits, states, strategies, and actions was proposed and further evaluated empirically in Studies 2 and 3. The results implied that for adapting to user traits, statistical expertise is a relevant dimension that should be considered. Additionally, for adapting to user states different user intentions such as monitoring and analysis should be accounted for. These results were used to develop a taxonomy which adapts visualization recommendations to these (and other) factors. A preliminary attempt to validate the taxonomy in Study 4 tested its visualization recommendations with a group of experts. While the corresponding results were somewhat ambiguous overall, some aspects nevertheless supported the claim that a user-adaptive data visualization approach based on the principles outlined in the taxonomy can indeed be useful. While the present approach to user adaptivity is still in its infancy and should be extended (e.g., by testing more participants), the general approach appears to be very promising.
作为一个领域，数据分析目前正处于其发展的关键节点，因为在数据量不断增加、用户更加多样化以及自动化分析解决方案的背景下，出现了商品化现象，后者可能使专家分析师变得不再必要。本文的一个核心假设是，数据可视化应根据用户和上下文进行调整。这一想法最初在研究1中得到探讨，该研究表明，在一组专家自由选择可视化数据集的选项时，个体之间存在很大差异。为了为系统的分类方法奠定理论基础，提出了一个结合用户特征、状态、策略和行为的用户模型，并在研究2和研究3中进行了进一步的实证评估。结果表明，为了适应用户特征，统计专业知识是一个应考虑的相关维度。此外，为了适应用户状态，应考虑不同的用户意图，如监测和分析。这些结果被用于开发一种分类法，该分类法根据这些（以及其他）因素调整可视化建议。在研究4中对分类法进行验证的初步尝试，用一组专家测试了其可视化建议。虽然相应的结果总体上有些模糊，但某些方面仍然支持这样的观点，即基于分类法中概述的原则的用户自适应数据可视化方法确实可能是有用的。虽然目前的用户自适应方法仍处于起步阶段，应加以扩展（例如，通过测试更多参与者），但总体方法似乎很有前景。
