GCN–LSTM Analysis of Spatiotemporal Evolution of Node Centrality in Tourism Flow Networks

Abstract

With the development of artificial intelligence and spatio-temporal big data technologies, the dynamic evolution characteristics of the tourism flow network and the spatial structure changes of its core nodes have become research hotspots. Based on the theory of complex networks, this paper constructs a tourism flow network covering mobile phone signaling, online platforms and traffic data, with a focus on discussing the spatio-temporal heterogeneous evolution mechanism of node centrality. By introducing AI models such as Graph Neural Network (GCN) and Long Short-Term Memory Network (LSTM), multi-scale recognition and dynamic prediction of core nodes in the tourism flow are achieved. The dataset contains 47 counties and 90 days of tourism flow data, covering 10 million signaling records, 5 million OTA data, and 3 million traffic data, processed at the daily level. We adopted a split scheme of 70% training set, 15% validation set and 15% test set for model training and evaluation. The experimental results show that the model has a prediction accuracy of 0.10 in RMSE and is superior to traditional benchmark methods (such as STGCN and DCRNN). The research also revealed the trend of centrality reconstruction of tourism flow nodes under different periods, holidays and external interventions. The research results have important theoretical and practical significance for improving the efficiency of regional tourism regulation and optimizing the layout of core nodes.