Design of a Personalized Travel Itinerary Generation System Using Deep Reinforcement Learning and Graph Neural Networks

Under the big data environment, the emergence of a large amount of travel information has brought about a serious “information overload” to travellers' travel path planning. Therefore, the recommendation of travel routes is born. Although the visual content of user-created photos can intuitively reflect the user’s travelling preferences, most of the existing path recommendation studies ignore or do not directly use the visual information in the pictures. The path recommendation method based on path sequence model, while improving the accuracy of path sequencing, also has problems such as data sparsity and ignoring user preferences, which affects the recommendation accuracy. Therefore, this paper establishes a user’s visual preference model based on user-generated images, and combines the user’s visual preference to study the travel path recommendation algorithm based on deep learning to provide travellers with high-precision personalized travel plans. This paper explores a personalised travel itinerary generation system based on deep reinforcement learning and graph neural networks. The user’s visual preference model is established through user-generated images, and combined with the user’s visual preference, a graph neural network and improved self-attention network fusion model is used to obtain sequence information from the user’s historical interaction behaviour, add the current target-related location information, and provide a high-precision personalised travel plan. The experimental results show that the model outperforms traditional methods in terms of recommendation accuracy (0.85) and user satisfaction (4.6), and is able to significantly improve the accuracy of personalised travel itinerary recommendations and user satisfaction. For example, the system’s recommendation accuracy and user ratings are generally higher for natural scenery and historical sites.