Virtual Only | Track 2 | Session 1. Data-driven Urban Insights in the Digital Age - People, Places and Patterns

Background: An increasing body of research highlights the strong correlation between the built environment and human perception. Street View lmages (SVl), with its panoramic perspective and spatial continuity, has emerged as a vital medium for quantitatively assessing urban environments. Advances in computer vision have enabled the extraction of morphological metrics (e.g., green view index) from SVl, fostering new paradigms for studying the links between environmental features and perceptual responses. Furthermore, Artificial intelligence (Al) technologies have accelerated the transition from single-metric analyses to multidimensional perception modeling, offering new pathways to comprehensively understand how urban environments influence human perception. Research Objective: This study presents a systematic review of visual data applications in urban environment perception research from 2010 to 2025.lt focuses on three core objectives: (1) To systematically synthesize the research progress on the association mechanisms between physical environmental elements and human perceptual responses; (2) To trace the evolution of visual data sources and processing workflows, including SVl preprocessing, perceptual data annotation, and analytical methods; (3) To explore the contributions and emerging trends of technologies such as Artificial intelligence(Al), Virtual Reality (VR), and digital twins in advancing urban environment perception research. Methods: A systematic literature search was conducted using keywords such as "street view images" and "urban environment perception" in the Web of Science and Scopus databases, targeting English-language publications from 2010 to 2025 and supplemented by high-impact journal sources. A multi-stage selection strategy was employed. Tools such as CiteSpace were used to perform bibliometric analyses of research themes, collaborative networks, and knowledge structures, resulting in a domain-specific knowledge map. Representative publications were further analyzed to identify key research hotspots and to trace the developmental trends in visual data applications within urban environment perception studies. Main Findings and Significance: CiteSpace analysis reveals the multidisciplinary nature and evolving focal points of urban environment perception research. A thorough review of the literature indicates that accelerated urbanization has driven a shift from long-term observational cycles toward shorter temporal responses. The research time scale has compressed from multi-year intervals to seasonal, and even daily, timeframes-raising new demands on the efficiency of data collection and processing. In response, the field demonstrates four emerging trends: (1) Deep learning-based perceptual modeling is increasingly replacing traditional statistical methods; (2) Generative Al techniques, such as style transfer and data augmentation, are improving the efficiency of visual data processing; (3) Virtual reality simulations of urban environments are being used to rapidly generate perceptual datasets; (4) The integration of visual data with human mobility trajectories is enabling dynamic analysis of perceptual preferences and behavioral responses.Despite technological advances, two major gaps persist: first, key non-visual sensory dimensions such as noise and odor-remain insufficiently integrated, limiting a holistic understanding of environmental perception variability; second, the application of VR/AR and digital twin technologies is largely limited to small-scale, experimental settings, lacking evaluation frameworks in real-world urban environments. Theoretical contribution: This review identifies critical research gaps and introduces a causal framework linking visual stimuli to perceptual responses, offering theoretical support for applying environmental psychology at the urban scale. lt also contributes to the development of "computable urban perception" as a novel interdisciplinary paradigm, bridging computer vision with environment-behavior studies. Practical implications: The findings provide actionable insights for community-scale urban regeneration, support data-driven smart city governance, and offer urban planners a foundation for designing evidence-based and context-sensitive strategies for sustainable urban transformation.

Urban villages represent complex spaces where historical contexts intersect with modern developments, characterized by multi-agent interactions. Traditional renewal approaches often adopt top-down methods, overlooking the genuine demands of local stakeholders such as residents. This study proposes a multi-agent game-based spatial function optimization model, using Laomenxi in Nanjing as a case example. Initially, multi-source data, including LBS signals, Points of Interest (POIs), and spatial environmental data, were integrated to identify ten typical user profiles and behavioral chains. These profiles were categorized into four core agent types: residents, tourists, developers, and government, enabling the simulation of their multi-dimensional spatial demands under varying renewal orientations.Subsequently, a composite potential assessment system was constructed, integrating social, economic, cultural, and ecological orientations across land parcels and facility points. Indicators were standardized and combined using the entropy weight method to generate an integrated renewal potential map. Utilizing the NetLogo platform, the model defined the spatial preference functions of the four agent types, simulating a dynamic multi-agent game around targeted plots. Agents prioritized high-potential neighboring parcels, with land-use conflicts resolved through a roulette-wheel selection mechanism. The iterative evolution of primary parcel functions, constrained by predefined land-use proportions and stabilized integrated utility values across the four orientations, generated multiple optimization solutions. Ultimately, Pareto-optimal functional layouts were achieved through multi-round solution efficacy comparisons.The results demonstrate the model’s efficacy in balancing multi-agent spatial preferences, dynamically restructuring land use, and fostering functional collaboration. Significant improvements were observed in community livability, consumer services, cultural integration, and ecological connectivity. This research provides a practical, data-driven modeling approach supporting multi-objective collaboration and enhancing overall spatial benefits in urban village renewal.

Under the dual pressures of rapid expansion and high-density development in megacities, how to enhance residents' well-being and consumption experience through public spaces has emerged as a new challenge in urban governance. The limited open green spaces in high-density cities serve multiple functions, including ecological, recreational, and consumption roles. In recent years, megacities like Shenzhen have continuously promoted the construction of a "Park City," with 1,320 parks built by the end of 2024. However, prominent issues remain, such as uneven distribution of commercial formats, insufficient consumption vitality, and unbalanced space utilization. Traditional evaluation methods often overlook real consumption behaviors and spatial differentiation, making it difficult to precisely support the refined upgrading of park spaces. How to identify optimization directions impartially and efficiently through data-driven approaches has become a core challenge urgently needing resolution for the sustainable development of high-density urban parks. This study focuses on 36 typical urban parks in Shenzhen, proposing a spatial analysis framework for park consumption landscapes based on social media big data and knowledge graphs. Using Python crawler technology, we systematically collected 15,622 real user reviews from Dianping and Amap (AutoNavi) between January and December 2024, covering consumption records of all formats, including catering, cultural creativity, parent-child activities, and sports. Additionally, a stratified sampling questionnaire was designed to cover people of different ages, genders, and residential areas, with 623 valid questionnaires recovered. The questionnaire content involved indicators such as consumption frequency, preferences, and facility satisfaction. Through natural language processing (NLP) technology, review texts were cleaned and preprocessed. First, the Latent Dirichlet Allocation (LDA) topic model was used for high-dimensional topic clustering of user review data, automatically identifying 11 major consumption themes, including cultural and creative retail, parent-child experience, and characteristic catering. Second, combining questionnaire data with attribute tags of review users, Multiple Correspondence Analysis (MCA) was adopted to reveal the characteristics and potential correlations of different groups in format consumption. Finally, knowledge graph technology was utilized to structure the multi-dimensional relationships of "user attributes-consumption behaviors-consumption facilities-evaluations," supporting network visualization and association rule mining. The study found significant spatial differentiation in the consumption formats of Shenzhen's urban parks. Parks in core urban districts such as Futian and Nanshan, with superior location conditions and activity diversity, have far higher attractiveness and facility richness than newly developed areas. Reviews of cultural creativity and parent-child formats account for more than 50%, becoming high-frequency consumption categories for citizens. In contrast, parks in some newly developed urban districts, such as Pingshan District, suffer from monotonous business formats and a scarcity of consumption scenarios, with consumption-related check-ins only 8% of those in core areas, failing to meet residents' growing diversified needs. Knowledge graph analysis further reveals that young groups aged 18–35 tend to prefer trendy "internet-famous" spots, characteristic cultural creative consumption, and light catering, with 45% of single consumption amounts falling within the range of 40–100 yuan. Family groups show higher willingness to consume parent-child entertainment facilities, and 63% of users overall express expectations for diversified consumption scenarios. This study provides a big data-driven refined analysis paradigm for the upgrading of consumption formats and spatial layout optimization in high-density urban parks like Shenzhen. It innovatively reveals the consumption network structure based on knowledge graphs, offering theoretical and practical references for the precision governance and multi-value enhancement of high-density urban parks globally.

a) Background In the digital age, the surge in high-resolution urban data is rapidly transforming urban life and economies, giving rise to 24-hour cities. These transformations require smarter planning tools to drive innovations in urban governance and collaborative decision-making. From night shift workers to early morning activities for seniors, modern urban lifestyles demand healthcare services and public spaces with spatial-temporal coverage, immediate response, and preventative care. Traditional static, time-limited healthcare facility layouts cannot meet these dynamic, all-day needs, especially at night. This creates a critical "sensing gap," as current digital sensing infrastructure ignores nighttime street-level realities. Most research ignores multimodal models like CLIP and fails to integrate micro-experience data (e.g., streetscapes and social semantics) for behavioral prediction or facility planning. This study proposes "24/7 Healthy Streets"—dynamic, responsive urban health lines—to support the continued vitality and resilience of 24-hour cities. b) Research Objectives and Questions This study aims to address the urgent need for more precise, equitable, and accessible health-support infrastructure at the community level. Streets are critical urban interfaces, and service quality must be continuously monitored across time and user types. This study introduces a novel evaluation model, StreetCLIP-GNN-LSTM, which combines multimodal image-text embedding (CLIP), graph neural networks (GNNs), and temporal modeling (LSTMs). Its goal is to enhance smart city planning by considering street-level visual perception as a core layer of urban intelligence. The model addresses blind spots in nighttime governance and promotes spatial equity through behavior-based accessibility assessment. By combining "urban artificial intelligence" and "health equity" frameworks, this study creates a new paradigm for assessing and predicting healthcare facility needs in all-weather urban environments. c) Methods and Data This study defines pedestrian blocks around Singapore MRT stations as supernodes, with nearby points of interest (POIs) forming a subgraph based on spatial proximity. Each POI node incorporates multimodal attributes from over 10,000 CLIP-embedded street view images, enabling fine-grained modeling of urban form and function. The StreetCLIP-GNN-LSTM model integrates urban traffic structure, neighborhood spatial layout, street imagery, and temporal behavior into a unified framework, enabling accurate street-segment-level facility analysis. By fusing spatial maps with semantic and temporal features, the model overcomes the limitations of single-dimensional urban models. Its core contribution lies in creating interpretable, relevant feature engineering that reflects real-world health infrastructure performance, not just algorithmic complexity. d) Key Findings and Significance Technically, the model leverages GNN scalability to generate localized adaptive responses. Its interpretable multimodal feature design supports counterfactual simulations (e.g., adding clinics, improving lighting, reducing obstructions), helping planners assess trade-offs and enhance decision confidence. By combining CLIP-based visual data with a spatial map of POIs, the model quantifies spatial factors (e.g., visual openness, vehicle encroachment) and generates scalable, data-driven insights. This transforms urban AI from black-box predictions to transparent participatory planning, supporting the vision of healthy streets 24/7 and advancing spatial justice goals. Key breakthroughs lie in its street-scale resolution and lifestyle-centric approach. The model enables microscopic assessment and simulation of health-supportive environments for real-world scenarios such as nighttime emergency response, chronic disease monitoring, late-night social activities, and health risk reduction for night shift workers. These reflect time-sensitive, behavior-driven patterns often overlooked in static planning. By enhancing health security and promoting equitable access to healthcare infrastructure, the model supports a diverse population—night shift workers, socially active individuals, elderly, and vulnerable groups—thereby enhancing long-term urban resilience and economic vitality.

Amid accelerating global population aging, ensuring mobility equity and quality of life for the elderly has become a critical issue in urban planning and governance. At the same time, transformations in urban economic structures and the growing penetration of digital technologies into urban spaces and service systems are giving rise to new 24-hour city lifestyles. In this context, elderly mobility is not only essential to daily well-being but also serves as a key indicator of urban inclusiveness and digital responsiveness. In megacities, metro systems are a vital mode of transport for older adults, supporting diverse needs such as medical appointments, family care, shopping, and recreation. As “non-peak users,” older adults play an increasingly important role in sustaining urban vitality during off-peak hours and maintaining the continuous operation of 24-hour service ecosystems. However, most existing studies focus on commuter travel or walkability for seniors, with limited attention to elderly metro use and its relationship with the built environment, especially in high-density urban settings. With the widespread availability of smart card data and the rise of AI-powered analytical tools, it is now possible to capture elderly travel patterns at scale and uncover their spatial determinants, offering a data-driven foundation for more inclusive and adaptive urban planning. This study focuses on elderly metro riders in Shanghai, a city with the world’s most extensive metro system spanning over 800 kilometers and where over 23% of residents are aged 60 and above. Drawing on over 1.3 million smart card travel chains and employing interpretable machine learning(XAI) models, we identify key built environment factors shaping elderly metro destination choices, and propose typology-specific planning strategies from both global and local perspectives. Our findings show that elderly travel behavior is distinct from that of commuters, featuring lifestyle-oriented purposes and high sensitivity to spatial context. Metro destinations are strongly influenced by the presence of non-commuting facilities such as cultural venues, healthcare services, recreational spaces, as well as retail and daily service facilities that support active aging. In contrast, traditional indicators like density and land-use mix exhibit diminishing or even negative effects in saturated high-density areas, due to increased environmental complexity and cognitive load. Additionally, we observe that synergies among built environment elements, rather than the abundance of any single factor, play a decisive role, underscoring the importance of integrated, system-oriented planning approaches. This study highlights the value of identifying key built environment attributes surrounding both origin and destination stations to promote senior-friendly transit use. Planning interventions should prioritize functional synergy over isolated upgrades and account for spatial heterogeneity across different urban density profiles. By integrating large-scale smart card data with XAI models, this research not only provides methodological innovation but also yields actionable insights for building age-inclusive, real-time responsive, and transit-oriented urban environments.

Driven by both globalization and the internationalization of higher education, international students have become an important part of the urban demographic structure of many regions. However, existing urban planning and spatial governance focuses on improving the physical environment, while paying insufficient attention to communication barriers and the social responsiveness mechanisms faced by diverse groups such as international students. There is also a lack of quantitative evidence based on spatial differences and dynamic scenarios that affect these populations. Taking Tokyo as a case study, this study investigates communication barriers, demand distribution, and spatial mechanisms affecting international students in three typical spaces: campuses, cohabiting communities, and transportation hubs. The objective is to provide systematic evidence and propose optimization pathways for the creation of inclusive urban space. This study is based on a spatial-behavior-demand (SPB) framework and comprises three parts: First, spatial zoning and situational sampling were conducted, selecting typical educational institutions, residential communities, and transport hubs as representative scenarios. Twelve high-frequency interaction spaces—such as student welcoming zones, activity venues, and shared kitchens—were delineated to reflect the full-day, multi-node mobility paths in students' daily lives. Second, through three months of behavioral observation, video recordings, and in-depth multilingual interviews with international students. This process resulted in 127 valid cases of communication barriers and demand incidents, which were categorized into three main types: language barriers, cultural misunderstanding and social loneliness. Third, a database of obstacle events was established, associating incidents with spatial locations. Analyze the probability of the different types of barriers occurring in various spatial types and activity contexts, as well as their influencing factors. The findings showed that language barrier incidents were most frequent in transportation hubs and cohabiting communities. These incidents were concentrated around high-frequency interactions such as ticket purchase, seeking directions, kitchen use, and emergency situations. Cultural misunderstandings were most prevalent in social recruitment and course exchange scenarios on campus. Social loneliness was most pronounced in cohabiting communities and campuses, especially in cohabited kitchen scenes during the school year and holidays. Further regression analysis indicated that multilingualism, accessibility, openness of interaction area, and cultural facilities were negatively correlated with the probability of communication barriers. Spatial cluster analysis identified "hot spots" with high prevalence of communication barriers, mainly distributed in student activity areas, subway commuting nodes and public kitchens. In addition, the study found significant differences in barrier mitigation mechanisms across space types. Campus spaces relied on information identification and volunteer support, cohabiting communities focused on shared resource management and cultural interaction; and transport hubs used multilingual services and behavioral guidance to meet the immediate communication needs of international students in high pressure or temporary scenarios. The study shows that international students are not only recipients of 24/7 urban services, but also actively participate in urban life through digital technologies, self-centred communities, and temporal and spatial flexibility activities, and provide new demands and feedback on urban functions such as the night time economy, real-time transportation, and multilingual service systems. The innovation of this study lies in its use of multi-spatial, multi-situation, and quantitative data to reveal the spatial distribution and impact mechanisms of communication barriers for international students in typical urban spaces. It also clarifies the practical effects of key spatial elements in barrier mitigation. This study not only provides empirical evidence and optimization paths for inclusive urban construction in Tokyo, but also provides theoretical frameworks and practical experiences in spatial governance and public service optimization for cities facing similar multi-population challenges globally. The results of the study are of broad practical significance for global cities that are transforming into "24 hour connected cities".