Abstract - Quranic verse recommender systems are highly needed to assist users in understanding the content of the Quranic verses through natural language queries, where finding relevant verses remains challenging, especially for users who are not familiar with specific keywords or thematic contexts. This study proposes a novel approach, namely a hybrid embedding-based Quranic verse recommender system that combines semantic representations from Sentence-BERT (SBERT) and keyword-based representations using Term Frequency-Inverse Document Frequency (TF-IDF). The system does not rely on user interaction history, thus overcoming the cold-start problem and providing more accurate recommendations according to users’ explicit needs. Additionally, the system is equipped with a Large Language Model (LLM) based on LLaMA to generate contextual explanations that facilitate understanding of the recommended verses, thereby helping users gain deeper and spiritually relevant insights. Implementation is done through a Telegram bot, which makes it easy for users to access in daily interactions. System evaluation using expert reviews and user testing yielded a precision of 82.4% and an average user rating of 4.37 out of 5, demonstrating the effectiveness of the hybrid embedding approach combined with LLM in improving the accessibility and comprehension of Quranic texts digitally. This study makes a significant contribution to the development of AI-based recommender systems in the religious domain through a new approach that integrates embedding and LLM for natural language-based recommendations.
Keywords— recommender system, qur’anic verse recommender system, sentence-BERT, term frequency-inverse document frequency