Semi-automatic Construction of Bidirectional Dialogue Dataset for Dialogue-Based Reading Comprehension Tutoring System Using Generative AI

The goal of this paper is to semi-automatically construct a bidirectional reading comprehension dialogue dataset that enables bidirectional dialogue or debate on reading passages within a dialogue-based reading comprehension tutoring system. To achieve this goal, we developed a process for semi-automatically constructing bidirectional reading comprehension dialogue dataset. Using this process, ten English experts were able to construct 2,951 datasets, with an average difficulty level of 8.24 (high school level) and an average dialogue turn count of 9.75 per passage.