A Novel Context-Aware Travel Recommendation Model Based on Spatiotemporal Graph Convolution Neural Network

Public ratings of well-known tourist attractions often show a bias, with popular scenic spots receiving higher ratings. However, the existing travel recommendation algorithm typically considers factors like the geographical location, category and popularity of scenic spots while overlooking the deviation between the popularity of tourist spots and the actual viewing effect caused by changes in time. In this paper, a submodule called regional spatiotemporal graph convolutional neural networks (RSGCNN) is proposed. RSGCNN utilizes the regional spatiotemporal graph to model the interactions and dynamics of tourist behaviors over time, enabling more accurate recommendations that account for both spatial and temporal factors. It constructs a regional spatiotemporal graph with social significance for tourists and performs convolution operations on its weighted adjacency matrix to extract features. This method effectively handles various interactions between tourist groups. To operate the extracted graph features in the time dimension and generate accurate recommendations while ensuring training efficiency, we introduce gated expansive causal convolutional neural network (GECCNN). GECCNN incorporates a gating mechanism and expansive causal convolution to address the challenges of temporal sequence prediction, ensuring more reliable and efficient training for long-term recommendations. The model is validated using a scenic spot check-in dataset. Experimental results show that the proposed model outperforms basic methods and demonstrates further improvements when applied to datasets with abundant temporal information about scenic spots.