Abstract

This paper is a follow-up study to compensate for the limitations of the previous research on the application of the duplicate verification algorithm. MARC data was provided directly from M Library in Busan, and the duplicate verification algorithm of KERIS was implemented and applied in Python. We extracted pairs of records with matching book numbers and divided them into ‘same group’ and ‘mismatch group’ based on matching location symbols and volumes, and compared the results of the algorithm. As a result of applying the algorithm, 98.10% of the ‘same group’ and only 0.43% of the ‘mismatch group’ were determined to be the same material. By analyzing the duplicate record pairs that were determined to be mismatched as a result of the algorithm, we proposed three ways to improve the algorithm as follows. First, remove ISBNs that contain the phrase SET. Second, consider forward or backward matches as matches in the publisher category. Third, forward or backward matches for author entries were considered matches. As a result of the algorithmic improvements, the identical judgment increased to 98.29% in the same group, and the mismatch judgment increased from 93.40% to 93.63% with no change in the identical judgment in the mismatch group. This shows that the improvements can increase algorithm performance while suppressing the error of labeling different materials as duplicates.