NurQalby: An Emotion-Based Quranic Verse Mobile Application Recommender Using Hybrid Filtering and BERT

Abstract

Many Muslim individuals seeking Quranic guidance during emotional distress struggle to locate appropriate verses and du'a because existing applications rely on keyword-based search, creating a vocabulary mismatch with modern emotional expressions. This paper presents NurQalby, a cross-platform mobile application that recommends Quranic verses and du'a based on the user's detected emotion and selected situational cause. The system employs a three-stage hybrid pipeline: emotion classification using a fine-tuned DistilRoBERTa model, contextual filtering by emotion and cause, and cosine similarity ranking using BERT sentence embeddings. The mobile frontend is developed with Flutter and the backend with Python FastAPI. A curated dataset of approximately 100 Quranic verses and 70 du'a sourced from Kaggle underpins the system. The selected verses and du’a, together with their emotion and cause labels, were reviewed and verified by Islamic subject matter experts to ensure the contextual and spiritual appropriateness of the recommendations. Evaluation using Precision@k, Recall@k, and nDCG@k across 12 test queries yields Precision@3 = 0.8333 and nDCG@3 = 0.8245. Results are interpreted with acknowledgment that ground truth was rule-based, which inflates scores relative to independent human judgement.