任务驱动的可视化方法自动选择

Visualization techniques are widely used by many different disciplines for data exploration along with their rapid development. However, for a general user, a major obstacle in using visualization techniques is "selection difficulty". That is, it is challenging for them to determine which visualization representation is most suitable for revealing the data pattern. To overcome this problem, two studies are proposed in this project: (1) task-driven categorization of visualization techniques and (2) automatic design of task-dependent visualization techniques. Specifically, we propose the novel concept of task-driven automatic selection of visualization techniques. By decomposition, abstraction and stigmatization of classic data analysis tasks from different areas, we build a task based taxonomy of visualization representations. Taking multi-variable, graph and spatio-temporal data as examples, we investigate density field based visualization selection method and information theory based visualization optimization method. In addition, an automatic visual space-oriented coding algorithm of visualization is proposed. With the goal of intelligent visualization recommendations for any data, this project will facilitate the fusion of the theoretic analysis and the application study of visualization techniques. As such, visualization will become more popular and produce more values in industry.

可视化技术正从理论探索和方法创新，迅速走向各专业领域与大众应用。然而“选择困境”是当前阻碍普通用户体验和使用可视化技术的一大难题，因为用户不知道选择哪种可视表达最能有效呈现手中数据的特征。本项目围绕如何根据分析任务对可视化方法归类和如何基于任务实现可视化自动设计两个核心问题开展研究，提出任务驱动的可视化方法自动选择的概念，以常见多变量、图和时空数据为切入点，通过分解、抽象和归纳各领域典型分析任务，进一步提出基于密度场和信息论的可视化方法智能选择策略，提出面向视觉空间的可视化自动编码方法，实现分析任务与可视化方法的快速自动匹配和推荐和设计。本项目研究将促进可视化理论与应用研究的交叉融合，加速可视化技术的大众化、平民化进程，推动可视化技术产生更多更广的应用价值。

近年来，可视化技术正从理论探索和方法创新，迅速走向各专业领域与大众应用。然而选择困境是当前阻碍普通用户体验和使用可视化技术的一大难题。针对此问题，本项目围绕任务驱动的可视化方法分类和任务驱动的可视化设计两个方面深入开展了研究，提出了一系列解决方案，在基于元任务的可视化方法分类与比较、面向趋势分析的宽高比自动选取、面向类簇结构和物体边界的颜色自动生成、类间关系保持的大规模数据自动可视采样、面向多目标的图布局与高维数据降维等方面取得了重要突破，研制了高效的图表动画可视化系统，在国内外知名期刊和会议上发表论文32篇，其中在领域顶级国际期刊/国际会议（IEEE TVCG,IEEE VIS, ACM CHI）上发表高水平论文19余篇，并获得ACM CHI 2021最佳论文提名奖，申请发明专利4项，已授权专利3项，培养硕士博士研究生5人。

内容获取失败，请点击重试