Measuring urban security perception in transport from social media data to inform transport network models: A case study for Milan in 2019 and New York City in 2021

Urban security perception is a critical, yet understudied factor of the adequacy dimension of transport poverty. This study investigates the potential of geo-social media data to assess perceived security in urban transport networks across European cities. Using over one million geo-referenced X microblogging posts from Milan (2019) and 12 million from New York City (2021), we applied a multi-stage Natural Language Processing (NLP) pipeline, including emotion and sentiment analysis and a few-shot semantic classifier, to detect posts expressing fear and insecurity, particularly related to women’s urban mobility. These indicators were spatially aggregated and compared against reference datasets on security perception and reported crimes. Results reveal distinct spatial patterns: Milan exhibited centralised posting with sparse, artefact-prone peripheral signals, while New York City showed more dispersed clusters in areas like the Bronx, Brooklyn, and Queens. However, correlations with ground-truth security data were consistently weak (Pearson’s r < 0.07), limiting social media’s reliability as a proxy for city-wide perceptions. Supplementary analysis of geo-referenced Google Maps reviews in Milan identified higher relevance of perceived security concerns in specific place types, such as parks and public transport locations, but revealed low overall confidence scores using the social media-based models. While geo-social media offer scalable, low-cost insights, limitations in spatial granularity, platform-dependent posting behavior, and demographic bias underscore the need for refined methodologies and multimodal data integration in future research.