As large language models (LLMs) continue to advance and gain widespread use, establishing systematic and reliable evaluation methodologies for LLMs and vision-language models (VLMs) has become essential to ensure their real-world effectiveness and reliability. There have been some early explorations about the usability of LLMs for limited urban tasks, but a systematic and scalable evaluation benchmark is still lacking. The challenge in constructing a systematic evaluation benchmark for urban research lies in the diversity of urban data, the complexity of application scenarios and the highly dynamic nature of the urban environment. In this paper, we design CityBench, an interactive simulator based evaluation platform, as the first systematic benchmark for evaluating the capabilities of LLMs for diverse tasks in urban research. First, we build CityData to integrate the diverse urban data and CitySimu to simulate fine-grained urban dynamics. Based on CityData and CitySimu, we design 8 representative urban tasks in 2 categories of perception-understanding and decision-making as the CityBench. With extensive results from 30 well-known LLMs and VLMs in 13 cities around the world, we find that advanced LLMs and VLMs can achieve competitive performance in diverse urban tasks requiring commonsense and semantic understanding abilities, e.g., understanding the human dynamics and semantic inference of urban images. Meanwhile, they fail to solve the challenging urban tasks requiring professional knowledge and high-level numerical abilities, e.g., geospatial prediction and traffic control task. These findings provide critical insights for the effective utilization and further development of LLMs to advance urban-related tasks and research in the future.

Large language models(LLMs), with their powerful language generation and reasoning capabilities, have already achieved notable success in many domains, e.g., math and code generation. However, they often fall short when tackling real-life geospatial tasks within urban environments. This limitation stems from a lack of physical world knowledge and relevant data during training. To address this gap, we propose CityGPT, a systematic framework designed to enhance LLMs' understanding of urban space and improve their ability to solve the related urban tasks by integrating a city-scale 'world model' into the model. Firstly, we construct a diverse instruction tuning dataset, CityInstruction, for injecting urban knowledge into LLMs and effectively boosting their spatial reasoning capabilities. Using a combination of CityInstruction and open source general instruction data, we introduce a novel and easy-to-use self-weighted fine-tuning method (SWFT) to train various LLMs (including ChatGLM3-6B, Llama3-8B, and Qwen2.5-7B) to enhance their urban spatial capabilities without compromising, or even improving, their general abilities. Finally, to validate the effectiveness of our proposed framework, we develop a comprehensive text-based spatial benchmark CityEval for evaluating the performance of LLMs across a wide range of urban scenarios and geospatial tasks. Extensive evaluation results demonstrate that smaller LLMs trained with CityInstruction by SWFT method can achieve performance that is competitive with, and in some cases superior to, proprietary LLMs when assessed using CityEval. Our work highlights the potential for integrating spatial knowledge into LLMs, thereby expanding their spatial cognition abilities and applicability to the real-world physical environments. The dataset, benchmark, and source code are open-sourced and can be accessed through https://github.com/tsinghua-fib-lab/CityGPT.

Large language models (LLMs) have exhibited remarkable capabilities and achieved significant breakthroughs across various domains, leading to their widespread adoption in recent years. Building on this progress, we investigate their potential in the realm of local life services. In this study, we establish a comprehensive benchmark and systematically evaluate the performance of diverse LLMs across a wide range of tasks relevant to local life services. To further enhance their effectiveness, we explore two key approaches: model fine-tuning and agent-based workflows. Our findings reveal that even a relatively compact 7B model can attain performance levels comparable to a much larger 72B model, effectively balancing inference cost and model capability. This optimization greatly enhances the feasibility and efficiency of deploying LLMs in real-world online services, making them more practical and accessible for local life applications. Available resources are at https://github.com/tsinghua-fib-lab/LocalEval.

Urban spatio-temporal prediction is crucial for informed decision-making, such as traffic management, resource optimization, and emergency response. Despite remarkable breakthroughs in pretrained natural language models that enable one model to handle diverse tasks, a universal solution for spatio-temporal prediction remains challenging. Existing prediction approaches are typically tailored for specific spatio-temporal scenarios, requiring task-specific model designs and extensive domain-specific training data. In this study, we introduce UniST, a universal model designed for general urban spatio-temporal prediction across a wide range of scenarios. Inspired by large language models, UniST achieves success through: (i) utilizing diverse spatio-temporal data from different scenarios, (ii) effective pre-training to capture complex spatio-temporal dynamics, (iii) knowledge-guided prompts to enhance generalization capabilities. These designs together unlock the potential of building a universal model for various scenarios. Extensive experiments on more than 20 spatio-temporal scenarios, including grid-based data and graph-based data, demonstrate UniST’s efficacy in advancing state-of-the-art performance, especially in few-shot and zero-shot prediction.

Urban spatio-temporal prediction is crucial for informed decision-making, such as traffic management, resource optimization, and emergence response. Despite remarkable breakthroughs in pretrained natural language models that enable one model to handle diverse tasks, a universal solution for spatio-temporal prediction remains challenging. Existing prediction approaches are typically tailored for specific spatio-temporal scenarios, requiring task-specific model designs and extensive domain-specific training data. In this study, we introduce UniST, a universal model designed for general urban spatio-temporal prediction across a wide range of scenarios. Inspired by large language models, UniST achieves success through: (i) utilizing diverse spatio-temporal data, (ii) effective pre-training to capture complex spatio-temporal relationships, (iii) spatio-temporal knowledge-guided prompts to enhance generalization capabilities. These designs together unlock the potential of building a universal model for various scenarios. Extensive experiments on more than 20 spatio-temporal scenarios demonstrate UniST's efficacy in advancing state-of-the-art performance, especially in few-shot and zero-shot prediction. The datasets and code implementation are released on https://github.com/tsinghua-fib-lab/UniST.

The recent past has witnessed a notable surge in on-demand food delivery (OFD) services, offering delivery fulfillment within dozens of minutes after an order is placed. In OFD, pooling multiple orders for simultaneous delivery in real-time order assignment is a pivotal efficiency source, which may in turn extend delivery time. Constructing high-quality order pooling to harmonize platform efficiency with the experiences of consumers and couriers, is crucial to OFD platforms. However, the complexity and real-time nature of order assignment, making extensive calculations impractical, significantly limit the potential for order consolidation. Moreover, offline environment is frequently riddled with unknown factors, posing challenges for the platform's perceptibility and pooling decisions.

A considerable amount of mobility data has been accumulated due to the proliferation of location-based services. Nevertheless, compared with mobility data from transportation systems like the GPS module in taxis, this kind of data is commonly sparse in terms of individual trajectories in the sense that users do not access mobile services and contribute their data all the time. Consequently, the sparsity inevitably weakens the practical value of the data even if it has a high user penetration rate. To solve this problem, we propose a novel attentional neural network-based model, named AttnMove, to densify individual trajectories by recovering unobserved locations at a fine-grained spatial-temporal resolution. To tackle the challenges posed by sparsity, we design various intra- and inter- trajectory attention mechanisms to better model the mobility regularity of users and fully exploit the periodical pattern from long-term history. In addition, to guarantee the robustness of the generated trajectories to avoid harming downstream applications, we also exploit the Bayesian approximate neural network to estimate the uncertainty of each imputation. As a result, locations generated by the model with high uncertainty will be excluded. We evaluate our model on two real-world datasets, and extensive results demonstrate the performance gain compared with the state-of-the-art methods. In-depth analyses of each design of our model have been conducted to understand their contribution. We also show that, by providing high-quality mobility data, our model can benefit a variety of mobility-oriented downstream applications.

Traffic prediction is the cornerstone of intelligent transportation system. Accurate traffic forecasting is essential for the applications of smart cities, i.e., intelligent traffic management and urban planning. Although various methods are proposed for spatio-temporal modeling, they ignore the dynamic characteristics of correlations among locations on road network. Meanwhile, most Recurrent Neural Network based works are not efficient enough due to their recurrent operations. Additionally, there is a severe lack of fair comparison among different methods on the same datasets. To address the above challenges, in this article, we propose a novel traffic prediction framework, named Dynamic Graph Convolutional Recurrent Network (DGCRN). In DGCRN, hyper-networks are designed to leverage and extract dynamic characteristics from node attributes, while the parameters of dynamic filters are generated at each time step. We filter the node embeddings and then use them to generate dynamic graph, which is integrated with pre-defined static graph. As far as we know, we are first to employ a generation method to model fine topology of dynamic graph at each time step. Furthermore, to enhance efficiency and performance, we employ a training strategy for DGCRN by restricting the iteration number of decoder during forward and backward propagation. Finally, a reproducible standardized benchmark and a brand new representative traffic dataset are opened for fair comparison and further research. Extensive experiments on three datasets demonstrate that our model outperforms 15 baselines consistently. Source codes are available at https://github.com/tsinghua-fib-lab/Traffic-Benchmark .

Origin-destination (OD) flow data, which reflects population mobility patterns in the city, is very important in many urban applications, such as urban planning and public resource allocation, etc. However, due to the high cost of money and time during device deployment and social surveys, it is challenging to obtain OD flow data, especially in developing cities and emerging cities where historical OD flow data is scarce. Therefore, it is necessary to investigate a method that can generate OD flow in cities where OD flow data are not available. The research on modeling population mobility in the city has a long history. Traditional gravity models, etc., are too simple to model the complex population mobility; recently proposed machine learning models and deep learning models are not applicable in cities where data are scarce because the parameters must be fitted with abundant data. To solve the problem of difficult access to OD flow data, we propose a method to learn mobility knowledge with ample data in the source city and generate OD flow data in new cities namedGODDAG(GeneratingOrigin-Destination Flow viaDomainAdversarial Training). Our proposed method consists of two parts, one is a GNN (graph neural networks) based mobility model generating OD flow between every two regions based on regional attributes such as census and POI distribution, and the other is a domain adversarial training strategy to make the model have better transfer ability between different cities. Extensive experiments are conducted on two real-world datasets to prove the validity of our methods.

Origin-Destination (OD) flow contains the information of direction and volume of population mobility between different regions in a city, having significant value in public transportation resource allocation. In this paper, we explore population distribution to infer OD flows, which is called pop2flow (population distribution to OD flows) problem. Compared to the conventional OD forecasting problem by using the historical OD matrix, pop2flow is more challenging because the population distribution carries much less information. In order to solve the pop2flow problem, we proposed a model, Graph-based Spatial-temporal Embedding with Dynamic Fusion (GSTE-DF). Specifically, GSTE-DF is composed of two parts: node embedding learning and flow prediction. The node embedding learning part captures the dynamic spatial-temporal features of population distribution into each nodes embedding. The flow prediction part adopts the learned embeddings and POI (points of interesting) distribution of every two regions to infer the population interaction between them. By conducting extensive experiments on real-world datasets collected in Beijing and New York City, we demonstrate the superiority of GSTE-DF compared to state-of-the-art baselines

It is essential to predict crowd flow precisely in a city, which is practically partitioned into irregular regions based on road networks and functionality. However, prior works mainly focus on grid-based crowd flow prediction, where a city is divided into many regular grids. Although Convolutional Neural Netwok (CNN) is powerful to capture spatial dependence from grid-based Euclidean data, it fails to tackle non-Euclidean data, which reflect the correlations among irregular regions. Besides, prior works fail to jointly capture the hierarchical spatio-temporal dependence from both regular and irregular regions. Finally, the correlations among regions are time-varying and functionality-related. However, the combination of dynamic and semantic attributes of regions are ignored by related works. To address the above challenges, in this article, we propose a novel model to tackle the flow prediction task for irregular regions. First, we employ CNN and Graph Neural Network (GNN) to capture micro and macro spatial dependence among grid-based regions and irregular regions, respectively. Further, we think highly of the dynamic inter-region correlations and propose a location-aware and time-aware graph attention mechanism named Semantic Graph Attention Network (Semantic-GAT), based on dynamic node attribute embedding and multi-view graph reconstruction. Extensive experimental results based on two real-life datasets demonstrate that our model outperforms 10 baselines by reducing the prediction error around 8%.

Crowd flow prediction is of great importance in a wide range of applications from urban planning, traffic control to public safety. It aims at predicting the inflow (the traffic of crowds entering a region in a given time interval) and outflow (the traffic of crowds leaving a region for other places) of each region in the city with knowing the historical flow data. In this article, we propose DeepSTN+, a deep learning-based convolutional model, to predict crowd flows in the metropolis. First, DeepSTN+ employs the ConvPlus structure to model the long-range spatial dependence among crowd flows in different regions. Further, PoI distributions and time factor are combined to express the effect of location attributes to introduce prior knowledge of the crowd movements. Finally, we propose a temporal attention-based fusion mechanism to stabilize the training process, which further improves the performance. Extensive experimental results based on four real-life datasets demonstrate the superiority of our model, i.e., DeepSTN+ reduces the error of the crowd flow prediction by approximately 10%–21% compared with the state-of-the-art baselines.

Human mobility prediction is of great importance for a wide spectrum of location-based applications. However, predicting mobility is not trivial because of four challenges: 1) the complex sequential transition regularities exhibited with time-dependent and high-order nature; 2) the multi-level periodicity of human mobility; 3) the heterogeneity and sparsity of the collected trajectory data; and 4) the complicated semantic motivation behind the mobility. In this paper, we propose DeepMove, an attentional recurrent network for mobility prediction from lengthy and sparse trajectories. In DeepMove, we first design a multi-modal embedding recurrent neural network to capture the complicated sequential transitions by jointly embedding the multiple factors that govern human mobility. Then, we propose a historical attention model with two mechanisms to capture the multi-level periodicity in a principle way, which effectively utilizes the periodicity nature to augment the recurrent neural network for mobility prediction. Furthermore, we design a context adaptor to capture the semantic effects of Point-Of-Interest (POI)-based activity and temporal factor (e.g., dwell time). Finally, we use the multi-task framework to encourage the model to learn comprehensive motivations with mobility by introducing the task of the next activity type prediction and the next check-in time prediction. We perform experiments on four representative real-life mobility datasets, and extensive evaluation results demonstrate that our model outperforms the state-of-the-art models by more than 10 percent. Moreover, compared with the state-of-the-art neural network models, DeepMove provides intuitive explanations into the prediction and sheds light on interpretable mobility prediction.

Online services are playing critical roles in almost all aspects of users’ life. Users usually have multiple online identities (IDs) in different online services. In order to fuse the separated user data in multiple services for better business intelligence, it is critical for service providers to link online IDs belonging to the same user. On the other hand, the popularity of mobile networks and GPS-equipped smart devices have provided a generic way to link IDs, i.e., utilizing themobility tracesof IDs. However, linking IDs based on their mobility traces has been a challenging problem due to the highly heterogeneous, incomplete and noisy mobility data across services. In this paper, we proposeDPLink, an end-to-end deep learning based framework, to complete the user identity linkage task for heterogeneous mobility data collected from different services with different properties.DPLinkis made up by afeature extractorincluding a location encoder and a trajectory encoder to extract representative features from trajectory and acomparatorto compare and decide whether to link two trajectories as the same user. Particularly, we propose a pre-training strategy with a simple task to train theDPLinkmodel to overcome the training difficulties introduced by the highly heterogeneous nature of different source mobility data. Besides, we introduce a multi-modal embedding network and a co-attention mechanism inDPLinkto deal with the low-quality problem of mobility data. By conducting extensive experiments on two real-life ground-truth mobility datasets with eight baselines, we demonstrate thatDPLinkoutperforms the state-of-the-art solutions by more than 15 percent in terms of hit-precision. Moreover, it is expandable to add external geographical context data and works stably with heterogeneous noisy mobility traces.

Obtaining crowd flow distribution with recognized human intention is extremely valuable for a series of applications for metropolitan cities. Previous solutions look at spatial correlation and temporal periodicity based on historical crowd flow information to calculate future crowd flow distribution. However, these mechanisms cannot recognize the intention behind crowd flow. We address this problem by leveraging a key insight – people's intention behind their movement is highly correlated with the point-of-interest (POI) distribution of the corresponding regions and adjacent regions. Therefore, we proposeDeepFlowGento model the complicated relationship between crowd flow, POI, check-ins, and time to generate intention-aware crowd flow. Specifically, we solve the conflict between dynamic crowd flow and static POI distribution by fusing the information in both time and POI domains. Besides, we employ a sequence of residual blocks inDeepFlowGento address the challenges of modeling the diverse temporal rhythms and heterogeneous influence of POI. Furthermore, we examine the generated intention-aware crowd flow from two aspects to substantiate the reasonability ofDeepFlowGen. Extensive experiments demonstrate that our model outperforms the state-of-the-art solutions by at most 30 percent in terms of NRMSE of total crowd flow. Moreover, the correlation between the generated intention-aware crowd flow and the check-in distribution across different categories of POIs is as high as 0.90 and 0.80 in Beijing and Shanghai. Combined with extensive case studies, we demonstrate the strong ability of our model in generating intention-aware crowd flow.

Fine-grained population distribution data is of great importance for many applications, e.g., urban planning, traffic scheduling, epidemic modeling, and risk control. However, due to the limitations of data collection, including infrastructure density, user privacy, and business security, such fine-grained data is hard to collect and usually, only coarse-grained data is available. Thus, obtaining fine-grained population distribution from coarse-grained distribution becomes an important problem. To tackle this problem, existing methods mainly rely on sufficient fine-grained ground truth for training, which is not often available for the majority of cities. That limits the applications of these methods and brings the necessity to transfer knowledge between data-sufficient source cities to data-scarce target cities.

A large-scale system for obtaining fine-grained vehicle trajectories is becoming increasingly important because it lays a solid foundation for a wide range of downstream applications, such as urban traffic optimization, road network profiling, route planning, etc. Traditional methods recover the trajectories from GPS data from apps or coarse-grained traces collected from base stations, which are costly and, more importantly, only cover limited vehicles on the road. Thus, they are not applicable to downstream tasks. To fill this gap, we explore the possibility of recovering vehicle trajectories from the video data recorded by widely deployed traffic cameras. The major challenges lie in the quality of the captured image, low sampling rate, and unbalanced temporal and spatial distribution. To address these challenges, we propose a general system to recover vehicle trajectories at the level of the road intersection, where a novel iterative framework is developed to combine both vehicle clustering and trajectory recovery tasks, which improve their performance simultaneously. The key motivation is that vehicle clustering based on visual features can provide essential discrete points for trajectory recovery, while the recovered routes can introduce spatial-temporal constraints to the initial vehicle clusters for de-noising the false results and complement the missing results. To prove the feasibility of our framework, we collect and plan to release a city-scale traffic camera dataset consisting of 24 hours of videos from 673 cameras across 1,106 intersections. To the best of our knowledge, this benchmark is the first to contain the ground truth of vehicle trajectories with a wide range of spatial and temporal coverage in an urban environment. We conduct extensive experiments and analysis on datasets of different scales to demonstrate the robustness of our framework. Last but not least, we have already deployed the whole system in the business applications of SenseTime, China, including traffic signal control and traffic flow analysis. We highly expect this dataset to further facilitate the research in this field and contribute more to traffic optimization systems in the real world.

Crowd flow prediction is an essential task benefiting a wide range of applications for the transportation system and public safety. However, it is a challenging problem due to the complex spatio-temporal dependence and the complicated impact of urban structure on the crowd flow patterns. In this article, we propose a novel framework, 3- D imensional G raph C onvolution N etwork (3DGCN), to predict citywide crowd flow. We first model it as a dynamic spatio-temporal graph prediction problem, where each node represents a region with time-varying flows, and each edge represents the origin–destination (OD) flow between its corresponding regions. As such, OD flows among regions are treated as a proxy for the spatial interactions among regions. To tackle the complex spatio-temporal dependence, our proposed 3DGCN can model the correlation among graph spatial and temporal neighbors simultaneously. To learn and incorporate urban structures in crowd flow prediction, we design the GCN aggregator to be learned from both crowd flow prediction and region function inference at the same time. Extensive experiments with real-world datasets in two cities demonstrate that our model outperforms state-of-the-art baselines by 9.6%∼19.5% for the next-time-interval prediction.

Realistic simulation of a massive amount of human mobility data is of great use in epidemic spreading modeling and related health policy-making. Existing solutions for mobility simulation can be classified into two categories: model-based methods and model-free methods, which are both limited in generating high-quality mobility data due to the complicated transitions and complex regularities in human mobility. To solve this problem, we propose a model-free generative adversarial framework, which effectively integrates the domain knowledge of human mobility regularity utilized in the model-based methods. In the proposed framework, we design a novel self-attention based sequential modeling network as the generator to capture the complicated temporal transitions in human mobility. To augment the learning power of the generator with the advantages of model-based methods, we design an attention-based region network to introduce the prior knowledge of urban structure to generate a meaningful trajectory. As for the discriminator, we design a mobility regularity-aware loss to distinguish the generated trajectory. Finally, we utilize the mobility regularities of spatial continuity and temporal periodicity to pre-train the generator and discriminator to further accelerate the learning procedure. Extensive experiments on two real-life mobility datasets demonstrate that our framework outperforms seven state-of-the-art baselines significantly in terms of improving the quality of simulated mobility data by 35%. Furthermore, in the simulated spreading of COVID-19, synthetic data from our framework reduces MAPE from 5% ~ 10% (baseline performance) to 2%.

Smartphone mobile application (App) usage prediction, i.e., which Apps will be used next, is beneficial for user experience improvement. Through an in-depth analysis on a real-world dataset, we find that App usage is highly spatio-temporally correlated and personalized. Given the ability to model complex spatio-temporal contexts, we aim to apply deep learning to achieve high prediction accuracy. However, the personalization yields a problem: training one network for each individual suffers from data scarcity, yet training one deep neural network for all users often fails to uncover user preference. In this article, we propose a novel App usage prediction framework, named DeepApp , to achieve context-aware prediction via multi-task learning. To tackle the challenge of data scarcity, we train one general network for multiple users to share common patterns. To better utilize the spatio-temporal contexts, we supplement a location prediction task in the multi-task learning framework to learn spatio-temporal relations. As for the personalization, we add a user identification task to capture user preference. We evaluate DeepApp on the large-scale dataset by extensive experiments. Results demonstrate that DeepApp outperforms the start-of-the-art baseline by 6.44%.

Increasingly available city data and advanced learning techniques have empowered people to improve the efficiency of our city functions. Among them, improving urban transportation efficiency is one of the most prominent topics. Recent studies have proposed to use reinforcement learning (RL) for traffic signal control. Different from traditional transportation approaches which rely heavily on prior knowledge, RL can learn directly from the feedback. However, without a careful model design, existing RL methods typically take a long time to converge and the learned models may fail to adapt to new scenarios. For example, a model trained well for morning traffic may not work for the afternoon traffic because the traffic flow could be reversed, resulting in very different state representation. In this paper, we propose a novel design called FRAP, which is based on the intuitive principle of phase competition in traffic signal control: when two traffic signals conflict, priority should be given to one with larger traffic movement (i.e., higher demand). Through the phase competition modeling, our model achieves invariance to symmetrical cases such as flipping and rotation in traffic flow. By conducting comprehensive experiments, we demonstrate that our model finds better solutions than existing RL methods in the complicated all-phase selection problem, converges much faster during training, and achieves superior generalizability for different road structures and traffic conditions.

Map matching is important in many trajectory based applications like route optimization and traffic schedule, etc. As the widely used methods, Hidden Markov Model and its variants are well studied to provide accurate and efficient map matching service. However, HMM based methods fail to utilize the value of enormous trajectory big data, which are useful for the map matching task. Furthermore, with many following-up works, they are still easily influenced by the noisy records, which are very common in the real system. To solve these problems, we revisit the map matching task from the data perspective, and propose to utilize the great power of data to help solve these problems. We build a deep learning based model to utilize all the trajectory data for joint training and knowledge sharing. With the help of embedding techniques and sequence learning model with attention enhancement, our system does the map matching in the latent space, which is tolerant to the noise in the physical space. Extensive experiments demonstrate that our model outperforms the widely used HMM based methods more than 10% (absolute accuracy) and works robustly in the noisy settings in the meantime.

Online services are playing critical roles in almost all aspects of users' life. Users usually have multiple online identities (IDs) in different online services. In order to fuse the separated user data in multiple services for better business intelligence, it is critical for service providers to link online IDs belonging to the same user. On the other hand, the popularity of mobile networks and GPS-equipped smart devices have provided a generic way to link IDs, i.e., utilizing the mobility traces of IDs. However, linking IDs based on their mobility traces has been a challenging problem due to the highly heterogeneous, incomplete and noisy mobility data across services.

Human mobility prediction is of great importance for a wide spectrum of location-based applications. However, predicting mobility is not trivial because of three challenges: 1) the complex sequential transition regularities exhibited with time-dependent and high-order nature; 2) the multi-level periodicity of human mobility; and 3) the heterogeneity and sparsity of the collected trajectory data. In this paper, we propose DeepMove, an attentional recurrent network for mobility prediction from lengthy and sparse trajectories. In DeepMove, we first design a multi-modal embedding recurrent neural network to capture the complicated sequential transitions by jointly embedding the multiple factors that govern the human mobility. Then, we propose a historical attention model with two mechanisms to capture the multi-level periodicity in a principle way, which effectively utilizes the periodicity nature to augment the recurrent neural network for mobility prediction. We perform experiments on three representative real-life mobility datasets, and extensive evaluation results demonstrate that our model outperforms the state-of-the-art models by more than 10%. Moreover, compared with the state-of-the-art neural network models, DeepMove provides intuitive explanations into the prediction and sheds light on interpretable mobility prediction.