Machine Learning-Based Model for Visualization Memorability Prediction

Abstract

Creating impactful visualizations is essential for effectively conveying complex data insights. However, current visualization techniques often fall short in ensuring that the data remains memorable and easily understandable. This research addresses this problem by investigating the factors that influence the memorability of visualizations, with the purpose of enhancing how key insights are retained and utilized. The main hypothesis is that combining visual content with simple textual metadata can predict and improve the memorability of visualizations. Our study introduces a web-based system powered by deep learning to predict memorability, analyzing both visual elements and textual descriptors, such as recognizable objects. The model assigns a memorability score based on these inputs. Results show that the hybrid model, integrating image and metadata analysis, provides accurate memorability predictions. The system offers designers immediate feedback, enabling them to iteratively refine their visualizations. This data-driven approach supports the creation of more memorable visual content, enhancing information retention and impact.