More about HKUST
On Understanding Misleading Visualizations, Automatic Detection, And Prevention
The Hong Kong University of Science and Technology Department of Computer Science and Engineering PhD Thesis Defence Title: "On Understanding Misleading Visualizations, Automatic Detection, And Prevention" By Mr. Yu Ho LO Abstract: Visualizations such as charts and graphs are used as rhetorical devices in news, blogs, and social media to support and further strengthen the message conveyed to the audience. Research shows the persuasive power of visualization. With equivalent information, the presence of a visualization--compared to a table--has a significant effect on changing or reinforcing the reader's attitude toward the concerned matter. Misinformation on the Internet is a significant challenge to societies that rely heavily on digital information. Increasingly, charts are included in the dissemination of misinformation. While the study of misinformation on the Internet often focuses on textual content, it rarely addresses the associated visual graphics. This thesis examines the current state of the art in the automatic detection and prevention of misleading visualizations. The initial efforts in this field involved collecting misleading visualizations to confirm their existence. Researchers then conducted perceptual experiments to verify the misleading effects on human perception and reasoning. Automatic detection of misleading visualizations is a recent development in visualization research. This research explores the landscape of misleading visualizations by collecting over one thousand real-world examples and identifying 74 types of issues, forming a taxonomy of 12 categories. It also investigates the use of multimodal Large Language Models (LLMs) for detecting misleading visualizations, demonstrating their potential in analyzing complex charts and enhancing visualization literacy. Additionally, this work develops effective explanation methods, including Bret Victor's concept of Explorable Explanations, to educate audiences and guide authors in modifying misleading visualizations. Building on the groundwork of understanding, recent breakthroughs in detection methods make it possible to develop prevention mechanisms such as automatic correction suggestions and in-situ annotations. However, further development of this research direction requires benchmarks to compare the performance of different detection and correction methods. Date: Monday, 19 August 2024 Time: 10:00am - 12:00noon Venue: Room 3494 Lifts 25/26 Chairman: Prof. Jidong ZHAO (CIVL) Committee Members: Prof. Huamin QU (Supervisor) Prof. Chiew-Lan TAI Dr. Shuai WANG Prof. Hongbo FU (EMIA) Prof. Kwan-Liu MA (UC Davis)