ChartOptimiser: Task-driven Optimisation of Chart Designs (2504.10180v1)

Abstract: Effective chart design is essential for satisfying viewers' information needs, such as retrieving values from a chart or comparing two values. However, creating effective charts is challenging and time-consuming due to the large design space and the inter-dependencies between individual design parameters. To address this challenge, we propose ChartOptimiser -- a Bayesian approach for task-driven optimisation of charts, such as bar charts. At the core of ChartOptimiser is a novel objective function to automatically optimise an eight-dimensional design space combining four perceptual metrics: visual saliency, text legibility, colour preference, and white space ratio. Through empirical evaluation on 12 bar charts and four common analytical tasks -- finding the extreme value, retrieving a value, comparing two values, and computing a derived value -- we show that ChartOptimiser outperforms existing design baselines concerning task-solving ease, visual aesthetics, and chart clarity. We also discuss two practical applications of ChartOptimiser: generating charts for accessibility and content localisation. Taken together, ChartOptimiser opens up an exciting new research direction in automated chart design where charts are optimised for users' information needs, preferences, and contexts.