Towards Visualization Recommendation Systems

Abstract

Data visualization is often used as the first step while performing a variety of analytical tasks. With the advent of large, high-dimensional datasets and strong interest in data science, there is a need for tools that can support rapid visual analysis. In this paper we describe our vision for a new class of visualization recommendation systems that can automatically identify and interactively recommend visualizations relevant to an analytical task.

View recommendation for visual data exploration

“…Data visualization is perhaps the most widely used tool in a data analyst's toolbox [39]. Generally, generating a visualization involves specification of visualization type (e.g., bar chart, scatter plot), filtering (e.g., selecting subset of data to visualize), transformation (e.g., aggregation on data), parameters (e.g., assigning attributes to axes) and visual encodings (e.g., color, size).…”

“…Having user preference included in the recommendation scheme provides an important criteria to evaluate which visualizations are useful for different users [39]. At the same time these systems are targeted to keep the user interaction to the minimum so that users of all skill levels can use them effectively.…”

“…Scalability: Data-driven recommendation system can be computationally expensive as it involves generation and evaluation of huge number of visualization, therefore scalability becomes a challenge. Even for a moderate size dataset the exploration may involve exponential space of visualizations [39].…”

“…Different systems focus on various levels of coverage. It is important to understand how much of the space of potential visualizations is covered by the recommendation system [39].…”

“….24 & 5.25 , the target views are from refined query WHERE Age=='[80][81][82][83][84][85][86][87][88][89][90]', while the comparison views are from refined query WHERE Age=='[30][31][32][33][34][35][36][37][38][39][40]'. The age groups '[30-40)' and '[80-90)' are the 5 th and 6 th largest groups in the population of 10 Diabetes Dataset: Re-admissions for Different Age Groups Bar Charts groups as can be seen in Figure 5.11b.…”