Application of Geographic Information Artificial Intelligence in Spatio-Temporal Semantic Extraction

The explosive technological iterations in the current field of artificial intelligence have provided new ways of thinking and technical means for the analysis of spatial big data. Geographic Information Artificial Intelligence (GeoAI) is developing rapidly and has become a hot field of current research. This paper provides a detailed review of the cutting-edge research in GeoAI. Firstly, it clearly defines the scope of GeoAI research in the fields of geographic science and urban spatial science, and categorizes it into three types: symbolic GeoAI, connectionist GeoAI, and behaviorist GeoAI. Then, by introducing the concept of 'semantics' and based on the semantic types extracted by models, it further classifies these three logical approaches of GeoAI research, aiming to deeply explore the research questions, content, and technical methods involved in each classification. On this basis, the article summarizes current research focuses and proposes potentially promising future research directions in the era of large models. Through the study, it is found that symbolic GeoAI can effectively organize static geographic knowledge, dynamic geographic processes, and spatio-temporal urban knowledge; connectionist GeoAI focuses on spatial embedding-based semantic representation and prediction; while behaviorist GeoAI concentrates on assisting spatial plan generation. In practical applications, there are the largest number of connectionist GeoAI studies, which are often used in combination with the other two types of GeoAI in upstream and downstream processes. Overall, GeoAI is increasingly pursuing the integration of multimodal data sources; different types of GeoAI research may share common goals, increasingly requiring the combination of multiple modeling logics to solve problems; and the flexible use of large models as intelligent agents to help solve geographic problems has become a new feasible path.