Journal of Geography, Environment and Earth Science International

Recent advances in geospatial artificial intelligence (GeoAI) have highlighted the potential of combining natural language processing (NLP) and graph-based learning for extracting semantically enriched spatial knowledge. This study describes the development of a lightweight hybrid framework for context-aware platial information that integrates NLP techniques with Graph Neural Networks (GNNs) to extract and analyze geosemantic knowledge from textual and spatial data. The main purpose is to create a platial knowledge graph that represents cities not simply as locations on maps, but as multidimensional entities with spatial, cultural, historical, and social characteristics. The methodology applied is based on three levels. The first level involves the semantic analysis of descriptive texts about 163 Greek cities which contain historical, cultural and social characteristics. The second level consists in the extraction of geographic data, such as the location of cities and other natural features like mountains, rivers, and seacoast obtained from open sources such as OpenStreetMap. The third level involves the construction of a knowledge graph with multidimensional features and relations. More specifically, we construct a diverse platial knowledge graph where cities are represented as nodes connected to other cities based on semantic features derived from both linguistic content and spatial context. By integrating GNNs, we manage to model and predict latent relationships between urban entities, enabling clustering and comparison of cities with similar geographic and semantic profiles. The preceding methodology is applied to a dataset involving large Greek cities to discover whether geosemantic fusion can reveal patterns of urban identity and regional connectivity.

By adopting this methodology, groups of cities that present similar characteristics are identified, without necessarily being geographically adjacent. Indicative examples show that cities sharing similar semantic and environmental characteristics can be grouped even when they are geographically distant. For example, Ioannina and Tripoli can be grouped due to their mountainous character and historical importance, while cities such as Kalamata, Kavala and Rhodes, although geographically dispersed, are correlated through their coastal tourism profile. The study focuses on Greece but provides a generalizable model for the multidimensional conceptualisation of place, which can be extended to other regions or countries. The resulting embeddings and city groupings are evaluated through clustering analysis, enabling the identification of latent similarities in urban identity and regional characteristics. This research is an attempt to expand the scientific approach of Platial Information Science, proposing a flexible and scalable tool for analyzing cities in a way that goes beyond traditional geographic mapping. The results highlight the potential for applications in spatial humanities, cultural geography, and urban planning.