Shahrazad: Hybrid Arabic Book Recommendation System

Recommender systems have become central to digital content discovery, yet their application to Arabic literature remains underexplored. This paper presents Shahrazad, a hybrid Arabic book recommendation system that integrates collaborative filtering (CF) with SVD++, content-based filtering (CBF) using AraBERT-based sentence embeddings, and sentiment analysis with a fine-tuned AraBERT classifier trained on the LABR dataset. The framework leverages the complementary strengths of each component to enhance personalization and robustness in sparse interaction settings. Evaluation on held-out test data shows that the hybrid model achieves superior performance in top- N recommendation, with Precision@10 of 0.2028 and Recall@10 of 0.7262. To further improve recommendation quality, a diversity-aware reranking step using cosine similarity penalties was applied, reducing redundancy and increasing coverage. These results highlight the effectiveness of combining deep language understanding with collaborative modeling to advance Arabic book recommendation systems.