Same Facade, Different Roles: Network-Based Role Clustering of Real Estate Portfolio

Real estate portfolios are commonly organized through broad classifications that can conceal meaningful differences among individual residential units. This study develops a network-based diagnostic framework for identifying portfolio-specific quality roles and demonstrates it on 5,561 Dutch residential units. Units are analyzed separately by residential type and represented as cosine-similarity networks based on attributes. Random-walk, mean first-passage-time profiles are used to derive structural dissimilarities, which are transformed through a Gaussian kernel and clustered using spectral clustering. The resulting clusters are interpreted through systematic characterization of cluster profiles followed by conceptual labeling. Clear affinity blocks and strong within-building consistency indicate that the roles capture meaningful portfolio structure beyond conventional administrative classifications and broad geographic location. The study delivers a replicable procedure for translating multidimensional property data into interpretable portfolio roles, although the resulting labels describe relative positions within the analyzed portfolio rather than universal categories of Dutch housing.