A model-driven approach to automate data visualization in big data analytics

Abstract

In big data analytics, advanced analytic techniques operate on big datasets aimed at complementing the role of traditional OLAP for decision making. To enable companies to take benefit of these techniques despite the lack of in-house technical skills, the H2020 TOREADOR Project adopts a model-driven architecture for streamlining analysis processes, from data preparation to their visualization. In this article, we propose a new approach named SkyViz focused on the visualization area, in particular on (1) how to specify the user’s objectives and describe the dataset to be visualized, (2) how to translate this specification into a platform-independent visualization type, and (3) how to concretely implement this visualization type on the target execution platform. To support step (1), we define a visualization context based on seven prioritizable coordinates for assessing the user’s objectives and conceptually describing the data to be visualized. To automate step (2), we propose a skyline-based technique that translates a visualization context into a set of most suitable visualization types. Finally, to automate step (3), we propose a skyline-based technique that, with reference to a specific platform, finds the best bindings between the columns of the dataset and the graphical coordinates used by the visualization type chosen by the user. SkyViz can be transparently extended to include more visualization types on one hand, more visualization coordinates on the other. The article is completed by an evaluation of SkyViz based on a case study excerpted from the pilot applications of the TOREADOR Project.