Semi-Automated Dataset Generation for Residential Buildings Using Graph-Based Topological Modelling

Most of Italy’s residential building stock predates contemporary structural safety and energy efficiency regulatory frameworks. Today, policymakers face the challenge of choosing whether to prioritise renovation or opt for demolition and reconstruction; both options carry significant socio-economic and environmental consequences and require extensive knowledge of the built heritage. However, detailed architecture-specific data remain scarce, as existing databases lack granular information. Moreover, traditional urban-level knowledge mapping approaches may be resource-intensive. To address this data gap, this study proposes a semi-automated methodology for generating graph-based digital models representing residential building floor plans. Using graph theory, floor spatial layouts are mapped into connectivity graphs and transformed into topological models. These models are enriched with functional data about spaces by assigning conditional topological rules based on node centrality metrics. The method was tested on 98 buildings in Bologna, Italy, yielding an 89.8% success rate and demonstrating its effectiveness in data-limited contexts. The resulting dataset facilitates the analysis of floor spatial configurations and the extraction of geometric attributes, laying the foundation for future analyses that will integrate machine learning techniques for functional detection and typological clustering.