AI-Powered Restaurant Recommendation System with Personalized User Preferences Using Sentence-BERT and FAISS

This paper presents an intelligent restaurant recommendation system that combines content-based filtering withpersonalized user preferences using state-of-the-art natural language processing techniques. The system utilizes Sentence-BERTembeddings for semantic understanding and Facebook AI Similarity Search (FAISS) for efficient similarity computation. Ahybrid approach combining 70% text embeddings and 30% category features achieves superior recommendation quality.The system features secure JWT-based authentication, user dining history tracking with ratings, and personalized recommendations weighted by user preferences. Implemented using FastAPI backend and React frontend, the system processes 12,505 unique restaurants from the Zomato Bangalore dataset after applying multi-factor deduplication that reduced the original dataset by 75.8%. Experimental results demonstrate fast query response times (¡ 100ms for search, 45ms for content-based recommendations) and accurate personalized suggestions. The system addresses the cold-start problem inherent in collaborative filtering approaches while maintaining scalability and real-time performance