AI for urban well-being: using NLP to assess user sentiment in Dubai’s historic spaces

Purpose Historic urban districts worldwide face the challenge of balancing heritage preservation with contemporary infrastructure needs while maintaining the well-being of their residents. This study develops a context-aware natural language processing approach using bidirectional encoder representations from transformers (BERT) to analyse resident sentiment toward urban features in Dubai’s Al Ras district, a traditional commercial area experiencing rapid transformation. Design/methodology/approach Data were collected through structured field surveys with 85 participants and applied strategic data augmentation using OpenAI GPT-4. The augmentation process produced 1,247 additional responses, resulting in a dataset comprising 1,332 responses and 9,326 textual entries used for sentiment analysis. Among the four evaluated architectures (recurrent neural networks, convolutional neural network, transformer and BERT), our BERT model achieved an accuracy of 79.84% in predicting satisfaction ratings across three urban feature categories: physical infrastructure, visual aesthetics and social functionality. Findings Results reveal significant disparities in resident appreciation, with visual elements (historic facades and traditional architecture) receiving the highest positive ratings (3.0/5), social features averaging 2.7/5 and physical infrastructure scoring the lowest (2.5/5). Word frequency analysis identified “authentic,” “community” and “historic” as primary positive sentiment drivers, contrasting with negative terms “lost,” “chaotic” and “frustrating” associated with navigation and traffic issues. Originality/value This methodology provides urban planners with granular insights into resident experiences, enabling evidence-based interventions that preserve cultural heritage while addressing practical infrastructure concerns in rapidly evolving urban environments.