<p>Understanding spatial patterns among tourist attractions is critical for tourism planning and management. Social media review data offer valuable insights into spatial connections by capturing information on tourists’ perceived relationships and cognitive patterns. However, both the vast volume and unstructured nature of social media review data pose significant challenges for computational analysis and pattern identification. This study leverages the advanced natural language processing capabilities of large language models (LLMs) to systematically interpret social media reviews; combined with complex network theory and geographic information system (GIS) methodologies, it develops a comprehensive framework for analyzing spatial patterns among tourist attractions. By utilizing standardized prompts and structured output formats, LLMs can significantly improve the accuracy and scalability of social media review data analysis. The results of this study indicate that traditional urban historical landmarks remain central nodes in the tourism network, whereas newly developed urban areas are increasingly prominent as intermediate attractions. Analysis of the core-periphery hierarchical structure demonstrates that historical and cultural sites in the old city act as central nodes in the network, while newly developed urban areas are gradually seeing the emergence of intermediate attractions. These structural dynamics contribute to the strategic and efficient allocation of urban tourism resources. Community detection reveals variations in connections between different types of tourist attractions, providing essential support for differentiated approaches to cultural tourism development. The novelty of this study resides in the integration of LLMs with spatial analysis methods, which thereby offers a new approach to social perception analysis in tourism.</p>