Anomalous behavior detection plays a vital role in public safety, enabling timely identification and intervention of dangerous events in crowded places like metro stations. However, challenges such as severe occlusion, dispersed action areas, and the scarcity of abnormal behavior samples make reliable detection difficult. This paper focuses on fighting detection in metro scenarios and proposes DSVLM, a novel multimodal large-model-based approach. Built on InternVL2.5-8B, DSVLM enhances contextual understanding through a prompt fusion strategy and employs low-rank matrix adaptation to reduce data requirements during fine-tuning. Additionally, a dynamic image segmentation method improves the model’s ability to capture fine-grained details in high-resolution footage. Experiments conducted on a custom metro dataset show that DSVLM achieves 99.17% accuracy in recognizing normal behaviors while successfully detecting all fighting incidents, demonstrating its effectiveness. To address the issue of excessive visual token redundancy introduced by dynamic segmentation, we further propose PruneVLM, a training-free visual token pruning method. By combining binary search with a greedy algorithm, PruneVLM minimizes attention distribution deviation and accelerates inference. Experiments under various pruning ratios confirm that with up to 70% token reduction, detection accuracy remains comparable while inference speed nearly doubles, verifying the practicality and stability of the proposed solution.

Multivariate time-series forecasting holds immense value across diverse applications, requiring methods to effectively capture complex temporal and inter-variable dynamics. A key challenge lies in uncovering the intrinsic patterns that govern predictability, beyond conventional designs, focusing on network architectures to explore latent relationships or temporal dependencies. Inspired by signal decomposition, this paper posits that time series predictability is derived from periodic characteristics at different frequencies. Consequently, we propose a novel time series forecasting method based on multi-frequency reference series correlation analysis. Through spectral analysis on long-term training data, we identify dominant spectral components and their harmonics to design base-pattern reference series. Unlike signal decomposition, which represents the original series as a linear combination of basis signals, our method uses a transformer model to compute cross-attention between the original series and reference series, capturing essential features for forecasting. Experiments on major open and synthetic datasets show state-of-the-art performance. Furthermore, by focusing on attention with a small number of reference series rather than pairwise variable attention, our method ensures scalability and broad applicability. The source code is available at: https://github.com/yuliang555/MFRS

Real-time and accurate traffic prediction can help traffic managers better understand the future changing trend of traffic state, thus enabling them to better manage and plan transportation. However, due to the dynamic, nonlinear and periodic changes of traffic states, designing a prediction model presents significant challenges. This paper proposes a traffic prediction model called MCSTC which integrates multiple cycle features and spatio-temporal correlation to achieve more accurate predictions. The proposed model consists of three parallel branches, which are used to learn the recent changes, daily periodic changes and weekly periodic changes, respectively. Each branch is a spatio-temporal feature learning network that uses a spatial learning module and a temporal learning module to mine the spatio-temporal correlation information contained in traffic data. They are then fused through an adaptive fusion module. We evaluate MCSTC on several real-world traffic datasets of in different cities. Results show that the proposed MCSTC outperforms existing methods in both short-term and long-term predictions in all datasets.

Accurate traffic flow prediction can help traffic platform managers to dispatch resources, so as to better meet people's travel needs, and reduce traffic jams and accidents. This paper proposes a traffic flow prediction model so called DGNN based on dynamic graph feature learning. It consists of four parts: spatio-temporal embedding module, temporal convolution module, dynamic graph feature learning module and prediction module. The key idea of DGNN is to construct dynamic adjacency matrices through real-time traffic state information, then pass the matrices and the input sequence through a dynamic graph convolution network to extract dynamic spatial features. Experiments are established on three real-world public datasets with different data scale and number of sensor nodes, and compared with several baseline methods. The results show that DGNN outperforms baseline methods in both short term and long term traffic prediction tasks.

With the rapid growth of the number of private vehicles, searching for accessible parking spaces becomes intractable for drivers, especially during high-demand hours. In recent years, we are witnessing a number of sharing economy services. Contract parking sharing, as an innovative sharing economy mode, has the potential to alleviate the difficult parking issue and make full use of the urban parking resources. However, the uncertainties of both drivers’ parking demand and owners’ sharing supply make it challenging to achieve efficient sharing. Thanks to IoT technology, many current parking lots now record vehicles’ fine-grained parking data for billing purposes. Leveraging these fine-grained parking data, we exploit available contract parking spaces to share them with drivers that have temporary parking demand. Specifically, we propose $\mathrm{W^{2}}$ Parking, a w in- w in contract parking sharing system, which includes two key components: (i) an idle time prediction model to estimate available periods of parking spaces and (ii) a parking sharing model to schedule temporary users to have access to these available parking spaces under both demand and supply uncertainties using dynamic programming combined with a 2-approximation algorithm with performance-bound guarantees. we evaluate our system on seven-month real-world parking data from 368 parking lots with 14,704 parking spaces. Extensive experimental results show that our $\mathrm{W^{2}}$ Parking achieves more than 90% of accuracy in parking time prediction, and the utilization rate of contract parking spaces is improved by 35%.

Research grants have played an important role in seeding and promoting fundamental research projects worldwide. There is a growing demand for developing and delivering scientific influence analysis as a service on research grant repositories. Such analysis can provide insight on how research grants help foster new research collaborations, encourage cross-organization collaborations, influence new research trends, and identify technical leadership. This article presents the design and development of a grant-based scientific influence analysis service, coined as GIMPACT. It takes a graph-theoretic approach to design and develop the scientific influence analysis algorithms over a real research-grant repository with three original contributions. First, we model the scientific influence analysis problem as a graph-based analysis problem by constructing heterogeneous graphs from the grants dataset, including mining the dataset to identify and extract important features and represent such features as a research grants information network. Second, we develop the scientific influence analysis algorithms over the research grants information network, which compute the overall scientific influence score by integrating self-influence score and multiple co-influence scores. The self-influence score reflects the grant-based research collaborations among institutions, and the co-influence scores reflect various types of cross-institution collaborations in terms of disciplines and keywords (subject areas). Third, we leverage the cluster analysis on the institution graph as an example application of scientific influence analysis service. By partitioning the institution graph into K clusters, with K as one of the service interface parameters, we show how different disciplines and different keywords are co-related through the grant-based influence analysis. We evaluate GIMPACT using a real grants dataset, consisting of 2512 institutions and their grants received over a period of 14 years. Our experimental results show that the GIMPACT influence analysis approach can effectively identify the grant-based research collaboration groups and provide valuable insight on an in-depth understanding of the scientific influence of research grants on research programs, institution leadership, and future collaboration opportunities.

Predicting the Origin-Destination (OD) traffic is a fundamental problem and of great significance in transportation research and civil engineering. There are three expectations for a good OD traffic predictor: 1) higher accuracy; 2) longer horizon; 3) better applicability. This paper proposes a Hybrid Spatio-Temporal Network (HSTN) model to predict OD traffic. The model emphasizes capturing more comprehensive spatial correlations among the sources of the traffic flows and temporal correlations between historical values and future prediction. HSTN is designed to have a Hybrid Spatial Module (HSM) and a Hybrid Temporal Module (HTM). HSM consists of three units to learn three types of spatial relationships and HTM consists of two units to quantity the influence of the input sequence on the target result. We evaluate HSTN on three real-world datasets of different travel modes in different cities. Results show that the proposed HSTN outperforms existing methods in both short-term and long-term predictions in all datasets.

文中探讨了基于智能家居系统的家庭用电监测系统方案的设计.采用基于云服务架构的智能家居物联网系统架构,通过智能家居网关实现智能电器和智能电表用电数据的汇聚和对智能家电设备的远程控制能力.能耗管控业务作为云端业务部署,提供融合异构数据的数据总线能力.针对单一电器和电表数据,能耗监测模块支持独立数据采集和负荷分解算法,为全屋家电能耗监测管理提供精细化数据支持.

Exploring human mobility is essential for urban applications. To observe human mobility, various data-driven techniques based on different data sources, such as cell phone and transportation data, have been proposed. This paper investigates human mobility through the emerging vehicular Internet service on public bus system. The key idea is that if a passenger is using WiFi on the bus or his/her WiFi device is activated in the background, we know that the passenger is traveling on the bus. By fusing the network events generated by WiFi devices with the data from the automatic fare collection (AFC) system, and the bus GPS information, we exploit not only the origin but also the destination of a passenger. Based on this idea, we develop a novel system called ViFi-MobiScanner which consists of about 4, 800 mobile routers distributed in a city with 1, 992 KM 2 urban area. We develop an ID matching algorithm that matches part of the users' network identities and their smartcard identities anonymously. As a result, we have built a set of labeled samples with the reference of observation from smartcard data and use them to train a classifier to infer users mobility from their network activities. We evaluate ViFi-MobiScanner with both field tests and collected datasets associated with 168 million network events, 3.6 million trips, and 1.4 million users. The evaluation results show that ViFi-MobiScanner increases the observability on the passengers and trips by about 53.9% and 48.1% over the smartcard observations. ViFi-MobiScanner also helps to estimate the passengers' destination that cannot be observed by current smartcard systems and the estimation can be accomplished in minutes. Thus it expands the observability of mobility in object, temporal and spatial dimensions and provides unique insights on human mobility at metropolitan scales.

With the development of applications on end devices, such as cell phones and tablets, more and more passengers would like to have entertainment on these end devices when they are cruising on vehicles. Due to the limited computation ability of the end devices, some of these applications have back-end components on the edge clouds, which are realized by Service Entities (SEs). In this work, we propose a system named DSEP to Dynamically determine the SEPlacement, such that the maximum latency experienced by the passengers can be minimized. To this end, we first train two sequential neural networks to predict the position of each individual vehicle, and propose an efficient algorithm based on optimization relaxation and Lagrange decomposition to determine the SE placement. Through extensive real-data driven simulations, we find that with the two sequential neural networks proposed in this paper, there are less than 1 percent errors on estimating where the passengers will access the edge cloud system. When the computation resources in the edge cloud are limited, DSEP can reduce the response latency by up to 43 percent compared with the nearest placement scheme. Even averaging the performance improvement over all simulation settings, DSEP can reduce the response latency by 16 percent.

With a rapid growth of vehicles in modern cities, searching for a parking space becomes difficult for drivers especially in rush hours. To alleviate parking difficulties and make the most of urban parking resources, contract parking sharing services allow drivers to pay for parking under the consent of owners, reaching a win-win situation. Contract parking sharing services, however, have not yet been prevailingly adopted due to the dynamic parking time which leads to uncertainties for sharing. Thanks to the Internet of things technique, most of modern parking lots record vehicles' fine-grained parking data including entry and exit timestamps for billing purposes. Leveraging the parking data, we analyze and exploit available vacant contract parking spaces. We propose SParking, a shared contract parking system with a win-win data-driven scheduling. SParking consists of (i) a parking time prediction model to exploit reliable periods of free parking spaces and (ii) an optimal scheduling model to allocate free parking spaces to drivers. To verify the effectiveness of SParking, we evaluate our design on seven-month real-world parking data involved with 368 parking lots and 14,704 parking spaces in Wuhan, China. The experimental results show that SParking achieves more than 90% of accuracy in parking time prediction and the average utilization rate of contract parking spaces is improved by 35%.

Relation extraction is an important while challenging task in information extraction. We find that existing solutions can hardly extract correct relation when the sentence is long and complex or the firsthand trigger word does not show. Inspired by the idea of fusing more and deeper information, we present a new relation extraction method that involves the types of entities in the joint embedding, namely, Entity Type Embedding (ETE). An architecture of Recurrent Piecewise Residual Networks (RPRN) is also proposed to cooperate with the joint embedding so that the relation extractor acquires the latent representation underlying the context of a sentence. We validate our method by experiments on public data set of New York Times. Experiment results show that our method outperforms the state-of-the-art models.

Nowadays, most modern parking lots integrate IoT technologies such as license plate recognition, mobile payment and automatic entrance/exit control to improve the convenience of the drivers and parking lot managers. However when searching for a vacant parking space becomes normally difficult in a metropolitan area, a more advanced "Smart Parking System" that can predict vacancy, enable reservation and differentiate pricing is even more needed. In this sense, understanding the parking behavior of customers has great significance to the manager of the parking estate, the drivers and the urban administrations. This paper explores the patterns and predictability of differential parking flows. With the license plate recognition and mobile payment system deployed in the parking lots, we collect about 14 million parking records of more than 300 parking lots in a big metropolitan in China. The time of every vehicle entering and exiting the parking lot is recorded along with its plate number. We first extract some features from these records and use K-means cluster algorithm to categorize the vehicle-parking lot pairs into three clusters empirically. The three clusters of parking behaviors are interpreted as regular parking users, long time visiting users and short time visiting users. Secondly, based on the parking lot's historical occupancy patterns, we designed several methods to predict the occupancy of the parking lot for different types of parking. The occupancies from the three types of vehicles can be used for making differential pricing policies or reservation policies.

Text data is one of the dominating data types in Big Data driven services and applications. The performance of text classification largely depends on the quality of feature extraction over the text corpus. For supervised learning over text documents, the TF-IDF (Term Frequency-Inverse Document Frequency) weighting factor is one of the most frequently used features in text classification. In this paper, we address two known limitations of TF-IDF based feature extraction method: First, the conventional TF-IDF weighting factor lacks of consideration about the synonymous relationship between feature terms. Second, for big corpus with large number of text documents and large number of feature terms, the computational complexity of text classification increases with the dimensionality of the feature space. We address these problems by introducing an optimization technique based on the Inter-Category Distributions (ICD) of terms and the Inter-Category Distributions of documents. We call this new weighting factor TF-IDF-ICD, namely TF-IDF with Inter-Category Distributions. To further enhance the effectiveness of our TF-IDF-ICD method, we describe a TF-IDF-ICD threshold based Dimensionality Reduction (DR) optimization. We test the text classifier with a corpus of 10, 000 articles. The evaluation results show that the proposed TF-IDF-ICD based text classification method outperforms the conventional TF-IDF based classification solution by 7.84% at only about 43.19% of the training time used by the conventional TF-IDF based text classification methods.

Bus service is an important public transportation. Besides the major goal of carrying passengers around, providing a comfortable travel experience for passengers is also an important business consideration. Traditional fixed bus line planning and vehicle scheduling are difficult to meet the needs of the public due to the real-time changes in passenger flow. Opening new flexible bus lines during special periods becomes an effective measure to ease the pressure. The traditional method for designing new bus lines mainly rely on human experience and field investigations, which are both non-scalable and incomplete. The emerging Internet of Things (IoT) and Big Data technologies have provided us with new opportunities. In this paper, we use smart card fare collection systems' data and GPS tracing systems' data to predict the passenger flow and find potential flexible lines. At the same time, we actually operated the flexible lines and conducted a field verification of its effectiveness.

Face recognition has becoming an important and popular authentication technology for web services and mobile applications in recent years. The quality of facial obstruction removal is a critical component of face recognition, especially for mission critical applications such as facial recognition based authentication systems. It is well known that some facial obstructions may severely affect the extraction and recognition quality and accuracy of facial features, which in turn disturbs the prediction accuracy of facial recognition model and algorithms. In this paper, we propose a Facial Obstructions Removal Scheme (FORS) based on an Enhanced Cycle-Consistent Generative Adversarial Networks (ECGAN) for face recognition. By training a convolution neural network based facial image classifier, we identify those images that contain facial obstructions. Then the images with facial obstructions are processed by using the facial image converter of FORS and the ECGAN model, which removes facial obstructions seamlessly while preserving the facial features. Our experimental results show that the proposed FORS scheme improves the face recognition accuracy over some existing state of art approaches.

This paper introduces LogCloud, a cloud based logistics planning system. LogCloud is designed as Software as a Service (SaaS) running on cloud that meets the needs of various logistics businesses. It provides functions of solving individual vehicle routing problem (VRP) tasks online, storing pre-calculated solutions and analyzing cases similarity. By combining the abilities of fast computing, solution storing and offline analysis of cloud, LogCloud brings cost effective resource planning within logistics business. We develop and deploy LogCloud and test it in three practical logistics businesses. Results show that LogCloud has provided an easy-to-use solution with more cost-effective planning than experienced human operators.

Bus service is an important public transportation. Besides the major goal of carrying passengers around, providing accurate and reliable travel information for passengers is also an important business consideration. The route and its traveling time can directly affect the number of people choosing the line. Traditional approaches to obtain route and its traveling time rely on historical experience, which are both nonscalable and incomplete. The wide adoptions of GPS tracing systems in public transportation provide new opportunities. In this paper, we associate it with station locations to derive the consumed time between two stations, and make a short forecast. To our best knowledge, this is the first paper which utilizes bus GPS data to design route and give its consumed time for a passenger.

In recent years, with the development of Internet big data, the popularity of mobile terminals, and the extensive services of the LBS platform, mass-level trajectory data has been formed. Due to the large amount of data, various types, different sampling frequency, and storage of point sequences of spatio-temporal data, researchers often need to perform a series of processing to convert the original data into available data for trajectory mining and analysis. In this paper, we study the management of multi-scale trajectories and provide an ordered KNN query for it. In our paper, we first propose a data model based on trajectory segments(DMTS) for multi-scale trajectories. The model converts the trajectory from the original sampling point to a form of trajectory segment that is easy to understand and organize. Next, we propose a fast ordered KNN query based on DMTS, and verify the correctness and efficiency of the algorithm by several experiments. The innovations in this paper mainly include: (1) Propose DMTS-based organization and management for trajectory data, which will effectively improve the efficiency of trajectory mining and analysis. By dividing the trajectory into moving objects, point objects, and trajectory objects, the DMTS will be applied to multi-scale heterogeneous trajectory data. (2) A fast ordered KNN query based on DMTS is proposed. Firstly, we use shape-based compression to reduce the amount of data and complexity of calculation. Secondly, we approve the measurement of trajectory distance from point-to-point distance to point-to-segment projection distance, which improves the accuracy and efficiency of querying heterogeneous trajectory data.