Automatic Synthesis of Visualization Design Knowledge Bases

TL;DR

This work tackles the rigidity of fixed feature spaces in knowledge-based visualization design by proposing a data-driven pipeline that automatically synthesizes a knowledge base from a corpus of visualizations. The method extracts low-level design features, forms complex features through forward and backward selection, and renders the selected features into formal, auditable rules compatible with Draco 2. In benchmark tests, the synthesized KB achieves comparable or higher predictive accuracy than Draco 2 and demonstrates strong transfer by constructing a genomics visualization KB with up to 97% accuracy and positive expert feedback. The approach shows broad applicability across visualization domains (e.g., genomics) and offers configurability to incorporate domain-specific priors, enabling scalable, interpretable design reasoning and potentially guiding automated visualization authoring. Overall, the results validate automatic KB synthesis as a practical pathway to extend and tailor visualization knowledge bases beyond hand-authored constraints.

Abstract

Formal representations of the visualization design space, such as knowledge bases and graphs, consolidate design practices into a shared resource and enable automated reasoning and interpretable design recommendations. However, prior approaches typically depend on fixed, manually authored rules, making it difficult to build novel representations or extend them for different visualization domains. Instead, we propose data-driven methods that automatically synthesize visualization design knowledge bases. Specifically, our methods (1) extract candidate design features from a visualization corpus, (2) select features forward and backward, and (3) render the final knowledge base. In our benchmark evaluation compared to Draco 2, our synthesized knowledge base offers general and interpretable design features and improves the accuracy of predicting effective designs by 1-15% in varied training and test sets. When we apply our approach to genomics visualization, the synthesized knowledge base includes sensible features with accuracy up to 97%, demonstrating the applicability of our approach to other visualization domains.

Figures (15)

• Figure 1: How Draco 2 yang2023:draco2 expresses a visualization design and captures a feature.
• Figure 2: Enumerating initial features (key-value chains) from a declarative design spec. The identifiers in a Draco specification are replaced with numerical indices when converted to an abstract syntax tree.
• Figure 3: Conditions for combining initial features.
• Figure 4: Computing frequency vectors through the extraction steps.
• Figure 5: Un-grounding features by stripping out membership and parent information.