Advancing Quranic Semantics: BERT-Based Verse Relationship Classification

Classification of semantic relationships between Quran verses supports Islamic scholarship, yet Arabic’s complexity challenges traditional methods. Machine learning (ML) models achieve up to 80.65% F1, deep learning (DL) models reach 80.74% F1, and BERT-based models—AraBERT, QARiB, Quran BERT—demonstrate superior performance, with Quran BERT (6 epochs) attaining 94.0% F1 and 94.1% accuracy. Specialized preprocessing and contextual embeddings drive BERT’s success, enabling applications in digital Tafsir and thematic analysis.