A Context-Aware Neural Collaborative Filtering Framework for Personalized Travel Recommendation

The proliferation of digital tourism platforms creates severe choice overload for travelers. Traditional recommender systems—predominantly Matrix Factorization and Content-Based Filtering—fail to integrate dynamic travel context into the neural latent space, resorting to sub-optimal post-filtering. This paper proposes a Context-Aware Neural Collaborative Filtering (CA-NCF) architecture embedding Season, Budget, and Duration as dense vectors alongside User and Item embeddings, processed through a fine-tuned MLP. A hybrid re-ranking function combining neural scores with review popularity and GPS-based proximity boost (Haversine formula) further refines results. Deployed as a Flask web application with OSRM routing, Nominatim geocoding, and Overpass API hotel discovery, the system transitions recommendations into actionable itineraries. Evaluated on 550+ Indian destinations and 2,300+ user ratings, CA-NCF achieves a final MSE of 0.100 versus 0.191 for a baseline NCF, alongside cold-start resilience and sub-100ms CPU inference.