Café Recommendation System
thesis project
→
Summary
Developed a deep learning-based café recommendation system by fine-tuning BERT (bert-base-uncased) for contextual embedding and sentiment classification on 38,000 Yelp reviews. Constructed café profile embeddings via mean pooling and implemented content-based ranking using cosine similarity combined with sentiment score (hybrid scoring). Achieved 97% nDCG (Top-5) in recommendation evaluation using Average Similarity Top-K and nDCG.