HetSpot: Analyzing Tourist Spot Popularity with Heterogeneous Graph Neural Network

Hiromasa Yamanishi, Ling Xiao, Toshihiko Yamasaki

Abstract: Tourism spot popularity prediction forecasts the prosperity of tourism destinations using heterogeneous information. Understanding the popularity of a spot can aid city planning and tourism site renovations. However, previous works overlooked interpretability in this task. Moreover, they only relied on a small portion of information on tourism spots and did not fully exploit the potential of Graph Neural Networks (GNNs) in tourism spot popularity prediction. To address the aforementioned problems, we propose a novel Heterogeneous GNN model, which we call HetSpot. By representing multimodal tourism spot information as a heterogeneous graph, HetSpot captures diverse cross modal information effectively to predict spot popularity. Moreover, we introduce an interpretation method for post-hoc analysis. Experimental evaluations on a large-scale Japanese tourism spots dataset show the superior performance of HetSpot. It achieves a high correlation value of 0.82 between our prediction and the actual tourist spot popularity, surpassing state-of-the-art models. Furthermore, the qualitative explanation results align well with our intuition based on experimental findings. The code and URL links used to construct the dataset are available online at https://github.com/HiromasaYamanishi/HetSpot