Optimization of the Jaro-Winker and Levenshtein Algorithms for a Thesis Title Filtering System at the UMB Faculty of Engineering

Abstract

Thesis title screening is crucial to prevent duplication of scientific work in the Faculty of Engineering, Universitas Muhammadiyah Bengkulu.² This study examines the optimization of the similarity threshold for Jaro-Winkler and Levenshtein Distance algorithms using 50 dummy thesis title pairs.³ Binary evaluation shows that the Jaro-Winkler algorithm achieves optimal performance with an F1-score of 1.000 at a 0.85 threshold.³ Meanwhile, Levenshtein Distance achieves its optimal performance at a 0.70 threshold with an F1-score of 0.837.³ The Wilcoxon Signed-Rank test confirms a statistically significant difference in performance between the two algorithms with  . The title screening system is recommended to implement the Jaro-Winkler algorithm.³