Beyond Numbers: Creating Analogies to Enhance Data Comprehension and Communication with Generative AI

TL;DR

Data-rich content often lacks immediate meaning due to unfamiliar measurements. The authors introduce data analogies and AnalogyMate, a two-stage pipeline that uses GPT-3.5 for analogy design and Stable Diffusion for illustration to improve data comprehension and communication. They derive a design space from 138 annotated cases, develop a two-stage creative pipeline, and validate the approach with a design-study and a crowdsourced comprehension study, showing increased ideation and engagement, though with some accuracy-related limitations. The work demonstrates a practical, flexible tool for data visualization and suggests future personalization and integrated illustration output to broaden impact.

Abstract

Unfamiliar measurements usually hinder readers from grasping the scale of the numerical data, understanding the content, and feeling engaged with the context. To enhance data comprehension and communication, we leverage analogies to bridge the gap between abstract data and familiar measurements. In this work, we first conduct semi-structured interviews with design experts to identify design problems and summarize design considerations. Then, we collect an analogy dataset of 138 cases from various online sources. Based on the collected dataset, we characterize a design space for creating data analogies. Next, we build a prototype system, AnalogyMate, that automatically suggests data analogies, their corresponding design solutions, and generated visual representations powered by generative AI. The study results show the usefulness of AnalogyMate in aiding the creation process of data analogies and the effectiveness of data analogy in enhancing data comprehension and communication.

Figures (7)

• Figure 1: The research process of the paper, where the blue boxes are different research activities while the green boxes are the output of such activities.
• Figure 2: Data analogy design space, which consists of two perspectives: data analogy design (analogy strategy, measurement transformation, and data binding type) and data analogy presentation (presentation form and layout).
• Figure 3: Four data analogy examples from our collected dataset. (a) 1.3 billion bottles are sold daily compared with the Eiffel Tower. (b) Five grams of plastic are consumed weekly compared to a porcelain soup spoon. (c) The wealth difference between Jeff Bezos and an American mid-class family compared with the size difference between a white blood cell and a finback whale. (d) Each year, 75.6 trillion gallons of water are added to the ocean, which equals 114.4 million Olympic-size swimming pools.
• Figure 4: System interface (left) and illustration created by users using generated analogy and materials (right). The system in this screenshot has suggested an analogy list for the input "Every day 1.3 billion plastic bottles are sold around the world" and generated illustration design solutions for the chosen analogy.
• Figure 5: Pipeline of AnalogyMate. In the analogy design process, the system first generates analogy objects, then modifies the results, and finally calculates and polishes the analogy sentence. In the illustration design process, the system suggests illustration design solutions and then generates illustration materials according to the keywords selected. Users can make crucial decisions interactively at critical points throughout the process.