Visual analytics for time-dependent data

考虑在一段时间内总结专利收集的任务，以揭示知识领域的进展；根据不同的参数评估运动员成绩的演变；分析突发疾病，目的是了解流行病传播的机制和可能的控制方法；或监测环境以评估其演变或特定变量对生态系统的影响。这些应用程序背后的关键概念是捕捉一种现象随时间的演变。虽然一些可视化技术可以用于时间相关方面的分析，但大多数可视化技术只适用于一维时间序列数据。对于不能直接建模为时间序列的数据，例如非结构化文档集合和动态图表，迫切需要新技术来支持发现和表示随时间变化的趋势和模式。拟议的研究将开发分析非时间序列数据集的时间方面的新方法，旨在设计工具和技术，以支持对大量数据的时间依赖模式的理解和解释。适当地处理时间依赖性给以用户为中心的分析环境带来了一些挑战，特别是在分析多个并发或依赖模式时给用户带来的认知过载。拟议的研究将调查新的视觉隐喻和技术，利用机器自动化的方法来定义一个多层次的以模式为中心的探索过程，允许用户在不同的抽象和细化层次上导航和探索时变数据。我们不仅打算为持久数据提供解决方案，还打算为流数据提供解决方案，流数据是指不断产生并需要在接收时进行处理的数据。通过工业和学术合作，我们将在三个不同的领域测试提出的方法：非结构化时间相关数据；时变图;轨迹数据。将从本文提出的进步中受益的应用程序示例包括运动表现分析、文本收集趋势分析和环境监测，这些应用程序可能会影响和改善专家和普通用户的日常生活。

Consider the tasks of summarizing a patent collection over the course of time to reveal the advances in a knowledge domain; assessing the evolution of an athlete's performance according to different parameters; analyzing an outbreak disease with the intent to understand the mechanisms of an epidemic propagation and potentially how to control it; or monitoring the environment to evaluate its evolution or the impact of specific variables on an ecosystem.The key concept underlying these applications is capturing the evolution of a phenomenon over time. Although some visualization techniques can be used for the analysis of time-dependent aspects, most of them are only suitable for unidimensional time-series data. With data that cannot be directly modeled as time-series, such as unstructured document collections and dynamic graphs, there is an acute need for new techniques to support the discovery and representation of trends and patterns that evolve over time. The proposed research will develop new approaches for the analysis of temporal aspects of non-time-series datasets, aiming to devise tools and techniques to support the understanding and interpretation of time-dependent patterns on large volumes of data.Appropriately handling the time dependency imposes several challenges for user-centered analytics environments, especially the cognitive overload imposed to users on the analysis of multiple concurrent or dependent patterns. The proposed research will investigate new visual metaphors and techniques, taking advantage of machine-automated approaches to define a multilevel pattern-centered exploratory process that allows users to navigate and explore time-varying data at different levels of abstraction and refinement. We intend to deliver solutions not only for persistent data but also for streaming data, that is, data that is continuously produced and need to be processed as received.With industrial and academic collaborations, we will test the proposed methods in three different domains: unstructured time-dependent data; time-varying graphs; and trajectory data. Examples of applications that will benefit from the advances proposed here include sports performance analysis, analysis of trends in text collections, and environmental monitoring, potentially impacting and improving the daily routines of both expert and casual users.