IMPLEMENTASI METODE HYBRID FUZZY JARO WINKLER DAN COSINE SIMILARITY PADA SISTEM PENCARIAN AYAT AL-QURAN BERBASIS TRANSLITERASI LATIN

This research addresses the challenge of retrieving Qur’anic verses in Latin transliteration, which is hindered by the absence of a standardized orthography, leading to diverse spelling variations. The study aims to design and implement a hybrid information retrieval system that integrates Fuzzy Jaro-Winkler for lexical similarity and Cosine Similarity on fine-tuned DistilBERT embeddings for semantic relevance. The system workflow begins with preprocessing and normalization of the dataset, followed by initial candidate selection using Jaro-Winkler, and final reranking through semantic similarity scoring. Evaluation was conducted using black-box testing across scenarios including ideal queries, spelling variations, incomplete queries, and varying query lengths. Results show high accuracy for ideal (96%) and varied spelling queries (92%), with performance improving as query length increases, reaching 96% for four-word queries. The hybrid approach effectively bridges lexical and semantic gaps, outperforming single-method baselines, and demonstrates robustness in handling non-standard transliteration in Qur’anic text retrieval.