Eye of the Beholder: Towards Measuring Visualization Complexity

TL;DR

This study investigates how to quantify visualization complexity by contrasting objective image-based metrics, manually coded feature models, and zero-shot LLM-based approaches against crowdsourced perception data. The crowdsourced ratings reveal that perceived complexity varies with familiarity and chart composition, while standard image complexity metrics show little predictive power. A multilinear regression using hand-coded features explains a portion of the variance (R^2 ≈ 0.49), highlighting which features raise or decrease perceived complexity; notably, legends ease understanding while complex compositions increase load. Importantly, a vision-enabled LLM (GPT4o-mini) can extract relevant features and closely approximate human complexity ratings with high consistency, offering a scalable alternative to manual coding. The work emphasizes the subjective nature of visualization complexity and demonstrates promising directions for scalable complexity assessment and visualization literacy enhancement.

Abstract

Constructing expressive and legible visualizations is a key activity for visualization designers. While numerous design guidelines exist, research on how specific graphical features affect perceived visual complexity remains limited. In this paper, we report on a crowdsourced study to collect human ratings of perceived complexity for diverse visualizations. Using these ratings as ground truth, we then evaluated three methods to estimate this perceived complexity: image analysis metrics, multilinear regression using manually coded visualization features, and automated feature extraction using a large language model (LLM). Image complexity metrics showed no correlation with human-perceived visualization complexity. Manual feature coding produced a reasonable predictive model but required substantial effort. In contrast, a zero-shot LLM (GPT-4o mini) demonstrated strong capabilities in both rating complexity and extracting relevant features. Our findings suggest that visualization complexity is truly in the eye of the beholder, yet can be effectively approximated using zero-shot LLM prompting, offering a scalable approach for evaluating the complexity of visualizations. The dataset and code for the study and data analysis can be found at https://osf.io/w85a4/

Figures (2)

• Figure 1: Complexity score plotted against GPT4o-mini's estimated complexity score. GPT4o-mini seems reasonably able to estimate the complexity of data visualizations, with some outliers.
• Figure 2: Complexity scores for (a) visualizations with and without legends, and (b) visualizations with and without annotations. Visualizations with a legend are on average perceived as less complex than visualization without one. Contrary to expectations, visualizations with annotations were perceived as more complex than those without annotations. This is likely a result of these annotations being hard to understand on their own, and therefore making the visualization seem more complex to the participants.