Public Space Quality Research Across Income Contexts: An NLP-Based Evidence Diagnosis Framework (1955–2026)
Year: 2026
Type: conference-paper
Abstract: AI-supported urban monitoring and digital decision-support systems increasingly depend on scientific evidence related to infrastructure, accessibility, safety, and environmental quality. Yet the knowledge base informing these systems remains uneven across world contexts. This paper proposes an NLP-based evidence diagnosis framework that combines TF-IDF, Sentence-BERT embeddings, Gaussian Mixture Model threshold calibration, and time-indexed thematic indicators to trace how public space quality research has evolved across income contexts from 1955 to 2026. The corpus contains 1,372 papers compiled from multiple scholarly sources, of which 1,017 were assigned to World Bank income groups based on the explicit empirical study setting reported in titles, abstracts, or keywords. Fourteen themes were operationalized through semantically curated descriptors, and theme activation was calibrated in both lexical and semantic spaces, with additional geographic and thematic validation procedures. To distinguish publication growth from thematic concentration, the analysis compares raw Weighted Context Intensity with its normalized counterpart. Results reveal a strong contextual asymmetry: high-income settings dominate the evidence base in both thematic breadth and temporal continuity, whereas lower-income settings remain sparse and discontinuous, especially for infrastructure-and equity-related topics. The framework therefore provides an interpretable diagnosis of contextual imbalance in the scientific record, with potential relevance for future context-sensitive urban monitoring applications.
Authors: Rym Ammar, Wadie Othmani
Language: English
Subfield: Library and Information Sciences
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7203596893