["Transactions in GIS, Volume 30, Issue 4, June 2026. ", "\nABSTRACT\nBuilding shape similarity measurement plays a crucial role in geospatial applications. However, this task remains challenging because hand‐crafted geometric encodings struggle to capture the full complexity of irregular building shapes. Spatial heterogeneity across regions undermines adaptability. Fully supervised approaches rely heavily on annotated datasets. To address these limitations, we propose a Spatial Context‐aware Weakly‐supervised Transfer learning (SCWT) framework. SCWT synergizes context‐enhanced multi‐view contour features with multi‐scale encoding to capture per‐vertex spatial contexts and global geometric semantics. Crucially, it introduces a weakly‐supervised paradigm that transfers knowledge from labeled simple shapes to complex ones, significantly reducing reliance on manual annotation while enhancing cross‐regional adaptability. Experimental results demonstrate that SCWT outperforms existing approaches in measuring building shape similarity. This framework offers a robust and scalable solution for shape analyses, advancing the automation of cross‐regional cartographic tasks and enabling intelligent integration of spatial data.\n"]