Museum Virtual Guide with Deep Learning-Based Hybrid Recommendations for Personalized User Experiences

Museums offer rich opportunities for cultural and educational engagement, but traditional exhibitions often lack personalization and interactivity. This paper presents a Museum Virtual Guide system, developed in collaboration with the National Museum of Egyptian Civilization (NMEC), that combines Augmented Reality (AR) with Deep Learning-Based Hybrid Recommendations to enhance user experience. For exhibit recognition, a YOLOv11 object detection model was trained on a museum dataset and integrated into Unity, allowing real-time detection and 3D model overlay through socket communication. Additionally, a deep hybrid recommendation model was developed, merging collaborative filtering and content-based approaches by leveraging exhibit text embeddings (BERT), image features (ResNet50), and categorical metadata. This hybrid design helps mitigate the cold start problem by incorporating content signals when user interaction data is sparse. The system predicts user preferences in a binary like/dislike format and delivers personalized exhibit suggestions. Experimental evaluations demonstrate high object detection accuracy (mAP@0.5 = 0.995) and effective recommendation performance. The proposed framework provides a scalable, immersive solution for personalized museum exploration.