Support for Reading Comprehension of Scholarly Papers in Japanese Using ChatGPT

Abstract

Given the prevalence of new concepts, technical terms, and the rise of e-book reading in scholarly papers, research has been conducted to enhance scholarly browsing by providing access to key terms, cited texts, and definitions of technical terms. We propose a method to extract various types of information from scholarly papers using ChatGPT and provide readers with helpful information for scholarly browsing. The proposed interface involves five key elements: a keyword tag cloud, a flowchart outlining the methodology, figure and table descriptions, a paper summary, and summaries of each section. These elements are presented to the reader alongside the PDF file of a paper. In our experiments, we measured the time required to answer questions about the paper's content and the percentage of correct responses to evaluate both the speed and depth of comprehension. The impact of the reading comprehension support interface on usability, understanding depth, and time to understand the paper was assessed via a questionnaire. Our experimental results demonstrated that while the interface did not reduce comprehension time, seven out of ten subjects reported slightly faster comprehension, with none experiencing slower comprehension. Half of the subjects indicated their understanding was somewhat deeper, and none reported a decline. Furthermore, nine out of ten subjects found the paper summary and section summaries to enhance their understanding.

Ähnliche Arbeiten