Multi-view Graph Attention Network for Travel Recommendation

As an e-commerce feature, the recommender system can enhance the consumer shopping experience and create huge benefits for businesses. The e-tourism has become one of the largest service industries with the application and popularity of recommender systems. Many studies have confirmed that the travel product recommendation is widely different from traditional recommendations. Due to the financial and time costs, travel products are usually browsed and purchased relatively infrequently compared with other traditional products (e.g., books, movies and food). In addition, choosing the appropriate travel product will be influenced by many factors, such as departure, destination and price. To tackle this challenging problem, we propose a MV-GAN (short for Multi View Graph Attention Network for travel recommendation) model. It enriches user and product semantics through both metapath-guided neighbors aggregation and multi-view fusion in heterogeneous travel product recommendation graph. In particular, we design node-level and path-level attention networks for learning user and product representations from every single view. To collaboratively integrate multiple types of relationships in different views, a view-level attention mechanism is proposed to aggregate the node representations and obtain global user and product representations. We evaluate the proposed method on a public dataset and a dataset constructed from a large tourism e-commerce website in China. Extensive experiments not only validate the effectiveness of MV-GAN, but also show its potentially good interpretability.

Automated summary

The recommender system plays a significant role in e-commerce and e-tourism, with travel product recommendation requiring consideration of multiple factors and differing from traditional recommendations. Researchers have proposed the MV-GAN model to enrich user and product semantics through multi-view fusion and neighbor aggregation, and experiments have shown its effectiveness and interpretability.