The integration of geo-spatial databases, advanced modeling, and artificial intelligence provides novel opportunities to investigate regional differentiation with direct implications for urban science and regional development. To avoid homogeneous competition and support regional tourism planning, it is important to examine how tourists perceive different tourism elements across large regional spaces. However, existing studies still lack a systematic workflow for extracting destination-specific perceptual characteristics and linking them to specific tourism locations. This study develops a UGC-based framework to identify tourism elements and perceptual characteristics, construct POI-level perceptual weights, and map the spatial patterns of tourism perception across Northeast China using 196,028 Ctrip reviews from 1491 POIs. The framework integrates jieba segmentation with BERT ranking for keyword extraction, ChatGPT-assisted filtering and manual review for keyword standardization, K-means clustering for perceptual dimension identification, and GIS-based kernel density analysis for spatial visualization. The main findings are as follows. First, tourism perception in Northeast China presented a diversified composite structure. Second, different perceptual dimensions and representative keywords showed differentiated spatial distributions. Third, the spatial structure of tourism perception in Northeast China was summarized as a composite spatial pattern centered on the Five-Dimensional Composite Perception Core Belt, supported by the Natural-Landscape Perception Belt and the Water Recreation Perception Area, and supplemented by multi-level urban nodes and inter-node perceptual linkages. This study provides methodological support and planning implications for understanding regional tourism differentiation and guiding evidence-based destination planning. The construction of geo-spatial databases plays a fundamental role in linking tourists’ perceptions to specific tourism elements and locations, enabling fine-grained and spatially explicit analysis that is difficult to achieve using conventional survey-based data.