2025 Volume 9 Issue 3 Pages e25-e31
This study aimed to improve the efficiency of metadata creation in digital archive construction. Using actual data from the "Kenzaburo Oe Library Digital Manuscript Archive," we reproduced the process of associating data sets through multiple methods and evaluated the consistency with expert-generated results. We compared and verified several methods, including exact bibliographic matching, string similarity comparison, and approaches utilizing large language models (LLMs). The results indicated that the methods using LLMs and exact matching of publication dates demonstrated a high accuracy rate. On the other hand, the method using LLMs faced the challenge of longer processing times. Based on these findings, we demonstrated that combining multiple methods can improve the balance between accuracy and processing time.