Screening the stones of Venice: Mapping social perceptions of cultural significance through graph-based semi-supervised classification

A novel approach using graph-based semi-supervised learning with graph neural networks (GNN) successfully mapped social perceptions of cultural significance across Venice. The method integrated multi-modal data—including visual and textual content from social media—and contextual connections such as spatial, temporal, and social factors into an attributed multi-graph structure. By training multiple GNN models as an ensemble, the study achieved high confidence in classifying cultural significance categories, which were then spatially mapped using message diffusion techniques on graphs. This framework not only provides a socially inclusive tool for heritage management but also demonstrates potential for broader applications in urban analysis, such as measuring safety, vitality, or popularity of spaces.

Bai, N., Nourian, P., Luo, R., Cheng, T., & Pereira Roders, A. (2023). Screening the stones of Venice: Mapping social perceptions of cultural significance through graph-based semi-supervised classification. ISPRS Journal of Photogrammetry and Remote Sensing, 203, 135–164. https://doi.org/10.1016/j.isprsjprs.2023.07.018