Abstract As cities adopt pervasive sensing, integrated data platforms, and AI, recommender systems are becoming central to shaping equitable, efficient, and citizen-focused urban services. This survey synthesizes peer-reviewed work across mobility, healthcare, energy, tourism, retail, and e-governance, offering a taxonomy linking collaborative filtering, content-based methods, hybrid designs, deep learning, and context-aware approaches to urban decision-making needs. Our review spans major smart-city domains, with most empirical studies in mobility and tourism, while deep and graph-based techniques remain unevenly distributed and comparatively rare in governance and energy. We examine how heterogeneous data sources, including IoT streams, geospatial signals, environmental indicators, and demographic attributes, are fused to support personalization under constraints such as latency, reliability, and privacy. The review highlights advances that address sparsity and cold start through graph neural models, sequence modeling, and transfer learning, and it covers operational enablers such as edge inference and streaming architectures for real-time recommendation. We assess risk and governance dimensions, including privacy preservation, fairness, exposure balance across neighborhoods, explainability, and mechanisms for audit and oversight. The survey identifies opportunities in pollution management, citizen education, and participatory platforms that broaden civic engagement. It also outlines how hybrid physical-virtual interactions, digital twins, immersive interfaces, generative models, and emerging quantum algorithms may reshape personalization and oversight in city-scale settings. Finally, we call for field evaluations and standardized benchmarks that jointly measure accuracy, latency, robustness to distribution shift, and equity.

Abstract Delineating the boundary or impact area of an economic, cultural, or lifestyle region has been a long-lasting problem in urban and regional geography. The fundamental difficulty lies in the exact definition of a region, and what criteria need to be considered. Existing methods either use criterion-based definitions or network-based measures to evaluate the affiliation of a city to a region. However, both types of methods only give static and definitive results but ignore the dynamism and graduality between regions. In this paper, we propose a Singular Value Decomposition (SVD)-based method to depict the impact areas of regions in China using individual connections among cities. Using the individual mobility data from an online map service, we decompose the mobility patterns of China into a series of eigen-mobility-patterns—each corresponds to the impact area of a city, or a mobility-based region. The overlay of multiple eigen-mobility patterns depicts the “blurred” boundary between the respective regions—or their competing hinterlands. We hope the method could be used to help understand the complexity of drawing regional boundaries and help policymakers to identify the non-confined but blurred economic and cultural landscape of various contexts in regional governance.

Abstract Urban emergencies significantly disrupt the subjective well-being (SWB) of urban population, while limited research has explored how time-use patterns interact with personality traits to shape SWB in crisis contexts. Understanding these mechanisms is essential for effective urban management and community resilience. This study investigates the influence of time use and personality profiles on SWB during urban emergencies, using data from the Shanghai lockdown. We identify three distinct personality profiles (i.e., Positive, Introverted, and Sensitive) and examine their heterogeneous responses. Our findings reveal that key quality-of-life factors, including health perception, social connection, and community liveability, directly influence SWB. Furthermore, time-use patterns, such as outdoor activities, paid work, sleep, online socialising, entertainment, and offline leisure, significantly affect residents’ life quality and SWB. In addition, personality traits moderate these effects: Positive individuals are particularly sensitive to sleep duration, while Sensitive individuals experience greater well-being variations due to outdoor activities. By revisiting the interactions between time use, personality traits, and SWB, our findings offer evidence-based guidance for policymakers and urban planners. This knowledge advances the understanding of psychological adaptation during urban emergencies and provides a foundation for more targeted approaches to community welfare, thereby strengthening community resilience during future crises.

This study investigates how the collective mobility (including movement and visiting) of residents and non-residents affects neighbourhood burglary levels. While past research has linked mobility to urban crime, this study explores how these relationships vary across population groups and social contexts at the neighbourhood level. Using mobile phone GPS data, we distinguished between residents and non-residents based on daily movement patterns. We then measured their mobility within defined spatial and temporal units. An explainable machine learning method (XGBoost and SHAP) was used to assess how mobility patterns influence burglary in London’s LSOAs from 2020 to 2021. Results show that increased collective mobility is generally associated with higher burglary levels. Specifically, non-resident footfall and residents’ stay-at-home time have a stronger influence than other variables like residents’ travelled distance. The impact also varies across neighbourhoods and shifts during periods of COVID-19 restrictions and relaxations. These findings confirm the dynamic link between mobility and crime, highlighting the value of understanding population-specific patterns to inform more targeted policing strategies.

Redlining is a discriminatory practice of systematically denying loans or mortgages to residents in specific neighborhoods based on racial or ethnical composition. In current literature research, there is a lack of understanding of the public perceptions of impacts of historical redlining practices at large geographic scales. Although some social groups and organizations conducted surveys or interviews to obtain public perceptions of it on small groups of people in certain areas, our knowledge of the impacts of redlining is limited and may reflect bias. This study used geotagged tweets from 2011 to 2023 to investigate public perceptions of redlining practices in U.S. counties. Multiscale geographically weighted regression (MGWR) was performed to explore both spatial heterogeneity and varying scales of associations between percentage of redlining-related geotagged tweets with negative sentiment and potential explanatory shaping factors in U.S. counties. Counties with a higher average household size, a higher percentage of people aged 45+, a lower homeownership rate, and a higher mobile home percentage have a significant association nationwide with more negative-sentiment expression in redlining-related tweets. However, counties with a lower insurance coverage are less likely to express negative sentiment in redlining-related tweets in some eastern U.S. counties, indicating a local significant association. The findings help people better understand the relationship between public perceptions of redlining practices and potential shaping factors. This study’s methodology can also be applied to investigate public perspectives or perceptions on other controversial social topics.

Abstract Crime prevention requires accurate prediction of the spatial and temporal distribution of criminal activities to effectively allocate law enforcement resources. However, many trending crime prediction algorithms lack comprehensive spatio-temporal structures and often consider only single input variables. This study innovatively using in ST-Cokriging method integrated both historical crime records as the primary variable and crime-related geo-tagged Twitter data as the co-variable for crime prediction. The predictive method has been specifically developed to assess crime risk across three major crime types—street crime, property crime, and vehicle crime—and applied in the San Francisco Bay Area (SFBA), California, a region characterized by high development and heightened crime sensitivity, for both prediction and validation. The results indicate that incorporating social media data into a spatio-temporal statistical method improves the associations between predicted and actual crime risk, reduced the Root Mean Squared Error (RMSE), and enhanced the identification of crime risk areas for both weekdays and weekends across three crime types compared to the method without the co-variable. This study presents a new multi-variable approach to more accurately predict crime, enabling law enforcement proactively address crime of varying nature in urban areas.

Abstract Prosperous waterway economics require rigorous safety measures. Unmanned aerial vehicle (UAV) offers massive images of inland waterways, within which navigation mark detection plays a critical role in ensuring waterway safety. This paper proposes a deep learning-based method for detecting navigation marks in UAV images. Firstly, a dataset of inland waterway navigation marks is constructed from UAV aerial images, which includes data collection, image enhancement, sample creation, and sample annotation. Secondly, a deep learning network model is developed, which uses ResNet-50 as the backbone, incorporates Coordinate Attention and Large-Scale Selective Kernel Attention mechanisms, integrates a Feature Pyramid Network (FPN) for feature enhancement, and uses Distance Intersection over Union (DIoU) as the loss function. Thirdly, the model is trained and evaluated on the constructed dataset, followed by precision assessment and post-processing. This paper explore a deep learning network model for small object detection in UAV images and establish a comprehensive workflow for detecting inland waterway navigation marks, thereby providing technical support for waterway safety.