Understanding Researchers' Behaviors and Design Considerations for AI-Assisted Scientific Caption Writing
Restricted (Penn State Only)
Author:
Ng, Ho Yin
Graduate Program:
Informatics
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
October 29, 2024
Committee Members:
Kenneth Huang, Thesis Advisor/Co-Advisor Saeed M Abdullah, Committee Member Dongwon Lee, Professor in Charge/Director of Graduate Studies Xiaolong (Luke) Zhang, Committee Member
Keywords:
User understanding AI writing assistant Caption writing Scientific writing
Abstract:
This study investigates the potential of AI, specifically Large Language Models (LLMs), in assisting researchers with figure caption writing—a crucial yet often tedious aspect of academic publishing. While previous research has focused on caption generation for readers, our study uniquely addresses the writer's perspective. We conducted a mixed-methods study with 18 participants, involving a writing task using AI-generated captions and semi-structured interviews. The study examined participants' caption writing practices, challenges, and views on AI assistance in scientific publishing. Quantitative analysis compared preferences among AI-generated caption and self-reported improvements, while qualitative analysis revealed insights across Task Characteristics, User Capabilities and Perceptions, Ecosystem Constraints, and Desired Interaction Features. Our findings provide a framework for understanding effective caption creation, inform researchers' writing processes, and identify design considerations to guide future AI-assisted academic writing tools in enhancing scientific communication.
