Cross-discipline is the growth point of new science and the inevitable trend of scientific development. The important application of location-based service, point−of−interest (POI) recommendation, as a research topic at the intersection of computer science and geographic information science, it plays an important role in promoting the cross−fertilization research of the two disciplines in the spatio−temporal data analysis and other related fields. This paper analyzes the influencing factors of POI recommendation, focuses on POI recommendation methods based on graph neural networks, including recommendation based on graph attention network, graph convolution network, and graph auto−encoder, with a comparative analysis of their respective characteristics; discusses some key challenges encountered in POI recommendation, such as data sparsity, cold start issues, and user dynamic preference issues, and proposes potential solution ideas to these challenges, makes prospects for the future development trend of POI recommendation.