GeoCaseRec: a knowledge graph semantics-driven framework for sustainable urban renewal recommendation

As urbanisation shifts from outward expansion to the regeneration of existing built environments, sustainable urban renewal increasingly depends on the effective reuse of historical cases. However, this remains constrained by fragmented multi-source geo-referenced data, semantic heterogeneity, and limited case evaluation. To address these challenges, this study proposes GeoCaseRec, a knowledge graph semantics-driven framework for urban renewal recommendation. Grounded in geographical case-based reasoning, the framework constructs an integrated representation model across four dimensions: geographical location, site conditions, spatial topology, and evolutionary characteristics. It further introduces a joint extraction method that combines large language models and spatial computation to derive key information from heterogeneous sources. Based on these data, intra-case spatial topologies and inter-case semantic hierarchies are constructed to automate an urban renewal knowledge graph. To bridge the technical gap in the recommendation phase, GeoCaseRec employs a Heterogeneous Graph Neural Network to fuse multi-source heterogeneous attributes, learn case embeddings and retrieve top-K similar historical cases to recommend potential evolutionary characteristics. Experiments on 474 projects across China’s four major urban agglomerations yield a semantic overlap score of 0.9277 and a top-3 precision of 0.7817, demonstrating the method’s utility for knowledge-based urban renewal decision support.