A hybrid expert-AI pipeline for constructing and characterising urban resilience indicator dependency networks

Characterising the interdependencies among urban resilience indicators is critical for effective governance, yet it remains constrained by the gap between qualitative policy constructs and quantitative system models. This paper presents a hybrid framework for constructing and characterising urban resilience dependency networks that integrates expert knowledge with transformer-based semantic inference. We synthesise 451 measurable sub-indicators from eleven global frameworks into the Urban System Abstraction Hierarchy (USAH), a reproducible graph dataset of 40 indicators across seven socio-economic domains, spanning institutional, economic, and social systems where earlier implementations stop at physical ones. Sentence-transformer embeddings project each indicator into a high-dimensional semantic space, and these are combined with an expert-derived prior to retain only dependencies supported by both semantic and operational evidence, yielding a directed dependency graph. Complex network analysis of the Vancouver case study shows a small-world topology, a modular structure in which governance-related indicators form a tightly coupled core, and a hub-sensitive robustness profile. Combining directional and positional centrality, the analysis assigns each indicator a functional role, anchor, bridge, peripheral driver, stabilizer, or receiver, distinguishing indicators that drive downstream domains from those that accumulate system state. The pipeline produces reproducible, internally consistent system representations that are transferable across cities and applicable to downstream policy analysis and network-based resilience modelling.