Visual Analytics at Scale: Supporting Bottom-up Explorations of Data

维多利亚大学、加拿大国防研究与发展和泰利斯集团建议开发可视化和创新的用户界面，以支持对海量、多维、时态数据集的自下而上的探索。在大规模--大数据--上可视化数据是信息可视化和可视化分析社区面临的需求。业界现有的工具往往侧重于自上而下的探索和分析，以支持趋势和模式发现。在这些情况下，用户可能对手头的数据有假设或明确的面向目标的问题，并可以指定对可用数据的高级查看和查询，以满足其信息需求。相比之下，自下而上的数据探索是一种跨许多信息域的需要，在这些信息域中，要分析的信息和文件可能事先不知道。这种情况经常发生在社交媒体分析、网络安全和取证等领域，分析师可能需要分析和演示工具来探索不同的、可能是动态的模式的陌生混合。尽管来自这些领域的分析师拥有很强的技术技能，但为勘探找到一个良好的起点往往是具有挑战性的。从我们对文献的回顾以及对可用解决方案的检查和试验来看，当前的视觉分析方法要么不能扩展到我们遇到的海量数据集，要么缺乏适当的认知支持和自下而上探索的视觉隐喻，要么缺乏协作支持。为了解决这一不足，我们的目标是开发创新的技术，以扩大和支持对大型和复杂数据集的自下而上的调查。我们将提供关于工具要求、设计工具原型的见解，并在工业环境中评估这些工具和要求。

The University of Victoria, Defence Research and Development Canada, and Thales Group propose to develop visualizations and innovative user interfaces to support bottom-up explorations of massive, multidimensional, temporal datasets. Visualizing data on a massive scale--Big Data--is a need facing the information visualization and visual analytics community. Existing tools from industry tend to focus on top-down exploration and analyses that supports trend and pattern discovery. In these cases, the user may have hypotheses or clear goal-oriented questions about the data at hand, and can specify high-level views and queries of the data available to meet their information needs. In contrast, bottom-up data exploration is a need that occurs across many information domains where the information and files to be analyzed may not be known a priori. This kind of situation occurs frequently in domains such as social media analysis, cyber security and forensics, where the analyst may require analysis and presentation tools to explore an unfamiliar mixture of different, possibly dynamic, schemas. Although the analysts from such domains have strong technical skills, finding a good starting point for the exploration is often challenging. From our review of the literature and examination and trials of available solutions, current visual analytic approaches either do not scale to the massive datasets we have encountered, lack appropriate cognitive support and visual metaphors for bottom-up explorations, or lack collaboration support. To address this shortfall, we aim to develop innovative techniques that will scale and support bottom-up investigations of large and complex datasets. We will provide insights on tool requirements, design tool prototypes, and evaluate these tools and requirements in industrial settings.