BIGDATA: F: DKA: Usable Multiple Scale Big Data Analytics through Interactive Visualization

从大数据中获得大的洞察力需要大的分析，这就带来了大的可用性问题。大数据分析通常依赖于几个计算和统计模型，这些模型在多个数据规模层面上运行，以发现和表征值得注意的模式。这些模型共同或依次对大数据进行过滤、分组、总结和可视化，以便分析人员对数据进行评估。作为大文本分析中的一个简单例子，首先对大量文本进行相关或代表性单词的采样，然后通过复杂的建模形式（例如，主题建模）进一步简化，然后通过应用降维算法实现可视化。随着数据规模的增加，模型的数量也在增加，同样地，分析过程中对人类交互的需求也在增加。通过互动，人类将专家判断纳入分析过程，并从不同角度有效地探索和理解大数据。然而，由于各种原因，与任何单个模型交互都是困难的，更不用说越来越多的模型了。因此，当前的人机交互研究与复杂的统计方法和快速计算相结合，以开发一个可用的、多模型的大数据分析框架。在软件的包装下，这个框架对专业用户和学生用户都可以访问；也就是说，可以在当前的政府和工业大数据集中做出新的发现，以及在新的教学模块中培养未来的本科和研究生水平的分析师。新的分析框架将视觉-参数交互（V2PI）扩展到多尺度V2PI （MV2PI）。V2PI目前支持可用的小数据分析，并允许用户通过直接与可视化中的数据交互来调整模型参数。也就是说，V2PI定量地解释可视化交互以更新底层模型参数并产生新的可视化。MV2PI现在将多个模型连接在一起，这些模型在统一的交互空间中以多个数据级别运行。在MV2PI中，可视化中的小规模数据交互传播到更大规模的模型（通过反转它们并更新它们的参数），并生成新的可视化。在文本分析示例中，如果用户将几个数据点拖到一起来假设一个集群，那么倒维降维模型将计算更新的维度权重，大规模查询相关的新命中，识别更改的主题，并更新布局以显示对新集群的大数据支持。有了MV2PI，用户可以实时交互地探索大规模数据和模型之间复杂的相互关系，并以一种可用的方式直接支持他们的自然认知语义过程。MV2PI的开发涉及：(1)制定明确的框架；(2)建立覆盖不同尺度、支持MV2PI模型反演的交互式模型（如交互式K-means和交互式潜Dirichlet Allocation）；(3)实现支持高性能、实时模型更新的计算方法；(4) MV2PI软件框架的可用性和有效性评估。项目网站（http://www.apps.stat.vt.edu/bava/mv2pi.html）将包括有关MV2PI开发、软件访问、数据集、教育材料和出版物的信息。

Gaining big insight from big data requires big analytics, which poses big usability problems. Analyses of big data often rely on several computational and statistical models that operate on multiple levels of data scale to discover and characterize noteworthy patterns. The models work jointly or in sequence to filter, group, summarize, and visualize big data so that analysts may assess the data. As a simple example in big text analytics, massive text is first sampled for relevant or representative words, then further reduced by a complex form of modeling (e.g., topic modeling), then visualized by applying a dimension reduction algorithm. As the size of data increases, so does the number of models and, likewise, the need for human interaction in the analytical process. By interacting, humans include expert judgment into the analytical process, and efficiently explore and make sense of big data from varying perspectives. However, for a variety of reasons, interacting with any individual model is difficult, let alone a growing number of models. Thus, current human-computer-interaction research is merged with complex statistical methods and fast computation to develop a usable, multi-model analytic framework for big data. Wrapped in software, the framework will be accessible to both professional and student users alike; i.e., available to make new discoveries in current government and industrial big datasets, as well as, educate future analysts at the undergraduate and graduate levels given new teaching modules. The new analytic framework extends Visual-to-Parametric Interaction (V2PI) to Multi-scale V2PI (MV2PI). V2PI currently supports usable small-data analytics, and enables users to adjust model parameters by interacting directly with data in visualizations. That is, V2PI interprets visual interactions quantitatively to update underlying model parameters and produce new visualizations. MV2PI now links together several models that operate at multiple levels of data-scale in a unified interactive space. In MV2PI, small-scale data interactions in visualizations propagate to larger scale models (by inverting them and updating their parameters) and new visualizations are generated. In the text analytics example, if users drag several data points together to hypothesize a cluster, the inverted dimension reduction model computes updated dimension weights, queries relevant new hits at the large scale, identifies changed topics, and updates the layout to show big-data support for the new cluster. With MV2PI, users may interactively explore large-scale data and complex inter-relationships between models in real time, and in a usable fashion that directly supports their natural cognitive sensemaking process. Development of MV2PI involves: (1) formulation of an explicitly stated framework ; (2) creation of new interactive models (e.g., Interactive K-means and Interactive Latent Dirichlet Allocation) that cover different levels of scale and support MV2PI model inversion; (3) implementation of computational methods to support high-performance, real-time model updates; and (4) evaluation of MV2PI software framework for usability and effectiveness. The project web site (http://www.apps.stat.vt.edu/bava/mv2pi.html) will include information on MV2PI development, access to software, datasets, educational materials, and publications.