KG4Vis: A Knowledge Graph-Based Approach for Visualization Recommendation.

Area of Science:

• Computer Science
• Data Visualization
• Artificial Intelligence

Background:

• Existing visualization recommendation systems often require manual rule specification or function as black boxes, hindering user adoption.
• General users, especially those without data visualization expertise, face challenges in creating effective visualizations.

Purpose of the Study:

• To present KG4Vis, a novel knowledge graph (KG)-based approach for explainable and effective data visualization recommendation.
• To overcome the limitations of existing rule-based and black-box machine learning methods in visualization recommendation.

Main Methods:

• Developed a framework for constructing knowledge graphs with entities (data features, columns, design choices) and their relations.
• Utilized a TransE-based embedding technique to learn KG embeddings from dataset-visualization pairs, capturing visualization rules.
• Inferred effective visualizations for new datasets using semantically meaningful rules from the KG.

Main Results:

• KG4Vis effectively models the mapping between data and visualizations without manual rule input.
• The approach provides explainable visualization recommendations, addressing the black-box issue.
• Extensive evaluations, including quantitative comparisons and case studies, confirmed the approach's effectiveness.

Conclusions:

• KG4Vis offers a viable solution for automatic and explainable visualization recommendation.
• The knowledge graph-based method lowers barriers for general users to create effective data visualizations.
• This approach enhances the accessibility and understandability of data visualization tools.