From Text to Insight: A Natural Language Processing framework for analysing community perceptions for urban climate resilience

Urban resilience strategies should reflect and strengthen community perceptions of environmental and climate risks. However, policymakers often struggle to engage diverse groups and incorporate insights from multiple and informal sources into resilience plans. This study presents a data-driven framework using Natural Language Processing to analyse community narratives regarding local risks and urban governance. By combining complementary analytical and visualisation techniques, we identify public concerns and perceptions and translate them into actionable urban resilience strategies. This approach is applied to a publicly owned neighbourhood in Lisbon facing aging infrastructure, limited access to services, energy poverty, and high unemployment. Primary data were collected from web-scraped sources and focus groups involving public authorities, local associations, and elderly inhabitants. Unsupervised topic modelling (BERTopic) was employed to identify key themes in digital narratives, including infrastructure quality, environmental conditions, and relations with local social institutions. Focus group transcripts were analysed through sentiment analysis, combining lexicon-based (VADER) and transformer-based (BART) models with expert validation and comparative assessment. Results show low levels of community awareness of environmental and climate risks and significant divergences between institutional and community perceptions of priorities, highlighting gaps in communication, institutional trust, and the perceived effectiveness of municipal renewal initiatives. In response, we propose soft co-design climate actions and targeted community awareness campaigns to strengthen risk perceptions and adaptive capacity. This framework offers transferable lessons for urban climate resilience by integrating insights from both digital and physical communities that are often underrepresented in formal decision-making, thereby supporting more inclusive local adaptation planning.