[Paper Review] Insight-centric Visualization Recommendation

Camille Harris, Ryan A. Rossi+6 more|arXiv (Cornell University)|Mar 21, 2021

Data Visualization and Analytics48 references10 citations

TL;DR

This paper introduces Insight-centric Visualization Recommendation, a novel approach that ranks and recommends both individual insights and higher-level insight-types (e.g., outliers, correlations) in data. By combining multiple learning-based methods to detect diverse insight types and ranking them for relevance, the system SpotLight enables users to quickly identify and explore the most important insights through annotated, visually grouped recommendations, significantly improving EDA efficiency and insight discovery in unfamiliar datasets.

ABSTRACT

Visualization recommendation systems simplify exploratory data analysis (EDA) and make understanding data more accessible to users of all skill levels by automatically generating visualizations for users to explore. However, most existing visualization recommendation systems focus on ranking all visualizations into a single list or set of groups based on particular attributes or encodings. This global ranking makes it difficult and time-consuming for users to find the most interesting or relevant insights. To address these limitations, we introduce a novel class of visualization recommendation systems that automatically rank and recommend both groups of related insights as well as the most important insights within each group. Our proposed approach combines results from many different learning-based methods to discover insights automatically. A key advantage is that this approach generalizes to a wide variety of attribute types such as categorical, numerical, and temporal, as well as complex non-trivial combinations of these different attribute types. To evaluate the effectiveness of our approach, we implemented a new insight-centric visualization recommendation system, SpotLight, which generates and ranks annotated visualizations to explain each insight. We conducted a user study with 12 participants and two datasets which showed that users are able to quickly understand and find relevant insights in unfamiliar data.

Motivation & Objective

• To address the inefficiency of existing visualization recommendation systems that rank all visualizations globally, making it hard to find the most relevant insights.
• To develop a system that automatically discovers, scores, and ranks both individual insights and higher-level insight-types (e.g., skew, multivariate outliers) in arbitrary datasets.
• To improve user efficiency in exploratory data analysis by organizing visualizations into ranked rows of insight-types, enabling faster identification of the most impactful insights.
• To evaluate the effectiveness of insight-centric recommendation through a user study with 12 participants on two datasets, assessing usability, insight discovery speed, and user satisfaction.

Proposed method

• The system uses a combination of supervised and unsupervised machine learning models to detect 21 distinct insight-types across categorical, numerical, and temporal attributes.
• Insight-types are scored and ranked globally based on the importance and relevance of the insights they contain, enabling prioritization of the most impactful insight categories.
• For each insight-type, individual insights are ranked by their significance, and appropriate visualizations are automatically generated and annotated for clarity.
• SpotLight groups visualizations into rows per insight-type, with rows ordered by insight-type relevance, allowing users to navigate by high-level insight categories.
• The system supports interactive filtering, bookmarking, and visualization consistency to enhance user interaction and exploration.
• A novel insight-type ranking mechanism is introduced, which scores and recommends the top-r insight-types based on the collective value of insights within each type.

Experimental results

Research questions

• RQ1 How can visualization recommendation systems better support users in quickly identifying the most relevant and impactful insights in unfamiliar datasets?
• RQ2 To what extent does ranking insight-types (as categories of insights) improve the efficiency and effectiveness of exploratory data analysis compared to ranking individual visualizations alone?
• RQ3 How do users perceive the usefulness, clarity, and diversity of visualizations and annotations generated by an insight-centric recommendation system?
• RQ4 What are the key usability challenges and desired enhancements for insight-centric visualization recommendation systems in real-world data exploration tasks?

Key findings

• Users were able to quickly understand and identify relevant insights in unfamiliar datasets, with 83% of bookmarked insights coming from the top-ranked insights within each insight-type.
• The system’s visualizations were rated highly (mean score of 5.5), especially due to the inclusion of annotations that significantly improved insight comprehension.
• Participants appreciated the diversity of chart types and found the system’s advanced visualizations more informative than those in existing tools.
• Users requested improvements such as more descriptive statistics, simplified explanations of statistical methods, and enhanced interactivity (e.g., zooming, details-on-demand).
• A key limitation identified was the reduced utility of insight-type ranking in datasets with fewer than 10 columns, suggesting greater need for such ranking in high-dimensional data.
• Users expressed a desire for greater control over visualizations, including customizing axes, binning, and color schemes, to support hypothesis-driven analysis beyond EDA.