2023, Vol. 4, Issue 2, Part A

There are many applications and uses of recommendation systems. For any recommendation system, the user-to-item interactions are important which can also be seen as graphs. We are focusing on travel or place recommendations. There are very few works on travel or trip recommendation systems using Graph Neural Networks (GNN) leveraging user-to- item interactions. In this work, we have implemented travel or place recommendations using the LightGCN model and compared it with the implementation of the traditional non-graph-based collaborative filtering Matrix Factorization (MF) approach. We have then shown that the LightGCN model performs better for travel or place recommendations than the Matrix Factorization approach. We achieved very good results [203% increase in precision, a 167% increase in recall, and a 98.6% NDCG increase in metrics] using a graph based LightGCN model for travel or place recommendations compared to the Matrix Factorization approach.