深度学习在城市计算中跨域数据融合的应用：分类、进展与展望

ZHONGHANG LI,South China University of Technology, ChinaLIANGHAO XIA,The University of Hong Kong, ChinaXUBIN REN,The University of Hong Kong, ChinaJIABIN TANG,The University of Hong Kong, ChinaTIANYI CHEN,City University of Hong Kong, ChinaYONG XU∗,South China University of Technology, ChinaCHAO HUANG∗,The University of Hong Kong, China Urban computing has emerged as a multidisciplinary field that harnesses data-driven technologies to addresschallenges and improve urban living. Traditional approaches, while beneficial, often face challenges withgeneralization, scalability, and contextual understanding. The advent of Large Language Models (LLMs) offerstransformative potential in this domain. This survey explores the intersection of LLMs and urban computing,emphasizing the impact of LLMs in processing and analyzing urban data, enhancing decision-making, andfostering citizen engagement. We provide a concise overview of the evolution and core technologies of LLMs.Additionally, we survey their applications across key urban domains, such as transportation, public safety,and environmental monitoring, summarizing essential tasks and prior works in various urban contexts, whilehighlighting LLMs’ functional roles and implementation patterns. Building on this, we propose potentialLLM-based solutions to address unresolved challenges. To facilitate in-depth research, we compile a list ofavailable datasets and tools applicable to diverse urban scenarios. Finally, we discuss the limitations of currentapproaches and outline future directions for advancing LLMs in urban computing. CCS Concepts:•Information systems→Data mining;•Computing methodologies→Artificialintelligence;Knowledge representation and reasoning. Additional Key Words and Phrases: Urban Computing, Large Language Models (LLMs), Spatio-Temporal DataMining, Transportation ACM Reference Format:Zhonghang Li, Lianghao Xia, Xubin Ren, Jiabin Tang, Tianyi Chen, Yong Xu, and Chao Huang. 2025. Urban Computing in the Era of Large Language Models. 1, 1, Article 1 (January 2025), 36 pages. https://doi.org/XXXXXXX.XXXXXXX 1Introduction In an era of rapid urbanization, cities worldwide face unprecedented challenges that stem fromincreasing population densities, resource constraints, and infrastructural demands [284]. UrbanarXiv:2504.02009v1 [cs.CY] 2 Apr 2025 computing emerges as a pivotal interdisciplinary field that harnesses the power of computingtechnologies to address these complex urban issues. By integrating data acquisition, analysis, andmodeling, urban computing endeavors to improve the quality of life in cities, enhance operationalefficiencies, and promote sustainable urban development [63,110,121]. It is effective across a rangeof practical urban scenarios. For instance, in transportation, it enables the optimization of trafficflow through intelligent traffic signal control and real-time routing suggestions, thereby reducingcongestion and emissions [41]. Environmental monitoring leverages sensors and data analytics totrack air and water quality, informing policy decisions and public health initiatives [266].Deep learning, known for its robust representation and relationship modeling, plays a key role in urban computing. Urban data, including traffic, safety, and environmental metrics, exhibitsstrong temporal and spatial correlations. To capture temporal dependencies, researchers frequentlyutilize models such as Recurrent Neural Networks (RNNs) [136], Temporal Convolutional Net-works (TCNs) [232], and attention mechanisms [182]. Graph Neural Networks (GNNs) are widelyused in spatial correlation analysis to model interactions between regions, enabling informationtransmission through message passing [108,111]. For tasks that require strategic decisions basedon various environmental conditions, such as traffic signal control, reinforcement learning is com-monly employed to optimize the decision-making process. Additionally, Convolutional NeuralNetworks (CNNs) are used in image recognition for tasks like land-use classification from satelliteimagery [14] and infrastructure anomaly detection [95].Despite these advancements, urban computing faced several bottlenecks that limited its full potential.i) Multimodal data processing capabilities.One significant challenge was the het-erogeneity and complexity of urban data. Urban environments generate vast amounts of data thatare diverse in nature, including numerical sensor readings, geospatial data, textual informationfrom social media, and unstructured data such as images and videos. Integrating and analyzingthis multimodal data to extract actionable insights proved difficult with traditional deep learningmodels, which often specialized in processing a single data type.ii) Generalization ability.Thegeneralization of deep learning models is limited by temporal and spatial distribution shifts in urbandata. Urban environments are diverse and dynamic, and models trained on historical data often failto adapt to new pat