Fine-Grained Identification of Renewal Problems in Older Residential Communities Using a Multimodal Large Language Model: A Case Study of Harbin, China
Abstract：To address the need for identifying micro-scale problems during the preliminary stage of older residential community renewal, this study develops a fine-grained identification framework based on a multimodal large language model, using 83 older residential communities and 70 310 valid street-view locations in Harbin, China, as the study sample. Spatial quality is assessed across six dimensions: building condition, pavement quality, parking and fire access, utility lines and cables, green infrastructure, and public facilities. Model-generated assessment rationales are standardized into “spatial object–problem state” combinations and aggregated to the community level for overall screening, dimension-specific weakness identification, and specific problem identification. The results show that the sampled communities exhibit moderate-to-low overall spatial quality, with utility lines and cables, building condition, and public facilities constituting the main weaknesses. Typical problems include dense and disordered overhead cabling, damaged and peeling fa?ades, missing public facilities, disorderly parking, and obstructed fire access. These problems exhibit either widespread distribution or local clustering. The proposed framework establishes an integrated scoring–semantic–spatial workflow and provides methodological support for large-sample screening, priority problem identification, and problem inventory generation during the preliminary stage of older residential community renewal.
