Contrastive learning make great success in multiple large language vision self-supervised tasks and gain the same attention in Multivariate Time Series (MTS) data analysis field. However, many existing MTS works focus on improving the representation of learning capabilities by a single supervised learning paradigm, inevitably ignoring the potential for jointly optimizing models across diverse supervised signals. To remedy the limitations, we propose a novel Class-Aware Spatio- Temporal Graph Contrasting Learning (CA-STGCL) framework, with a Multivariate Time Series Feature Extraction (MFE) Encoder-Decoder structure, and Contrastive Learning (CL) Decoder. Specifically, we first design an MFE Encoder-Decoder module to process the raw data and use the obtained representations to acquire different supervised signals. Then, we introduce an effective CL Decoder to obtain two distinct additional supervised signals by two data augmentation strategies and contrastive approaches. Further, we jointly optimize the model using three different supervised signals, i.e., the original representations signals by the MFE Encoder-Decoder module, and temporal and contextual contrasting signals from the CL Decoder. Finally, extensive experiments and comparisons are conducted on mul-tiple dominant MTS datasets, e.g., UCI-HAR, C-MAPSS, and ISRU C-S3. The results show that the proposed framework can surpass current promising methods and achieve the state-of-the-art performance11Code is available at: https://github.com/wyn404/CA-STGCL..

3D spatial data management is increasingly vital across various application scenarios, such as GIS, digital twins, human atlases, and tissue imaging. However, the inherent complexity of 3D spatial data, primarily represented by 3D geometries in real-world applications, hinders the efficient evaluation of spatial relationships through resource-intensive geometric computations. Geometric simplification algorithms have been developed to reduce the complexity of 3D representations, albeit at the cost of querying accuracy. Previous work has aimed to address precision loss by leveraging the spatial relationship between the simplified and original 3D object representations. However, this approach relied on specialized geometric simplification algorithms tailored to regions with specific criteria. In this paper, we introduce a novel approach to achieve highly efficient and accurate 3D spatial queries, incorporating geometric computation and simplification. We present a generalized progressive refinement methodology applicable to general geometric simplification algorithms, involving accurate querying of 3D geometry data using low-resolution representations and simplification extents quantified using Hausdorff distances at the facet level. Additionally, we propose techniques for calculating and storing Hausdorff distances efficiently. Extensive experimental evaluations validate the effectiveness of the proposed method which outperforms state-of-the-art systems by a factor of 4 while minimizing computational and storage overhead.

Spatial join has become a frequently used yet resource-intensive operation in geospatial applications, driven by the increasing volume and complexity of geospatial data. With Hadoop and Spark becoming the de facto standard platforms for distributed computing, scalable spatial data processing is primarily achieved by partitioning the input space to form parallel units on these platforms. Effective spatial data partitioning is critical for task parallelization and load balancing, but it faces significant challenges due to data skew and the geometric and topological complexity of spatial objects, particularly in supporting spatial joins. This paper examines the interplay among query performance, spatial data partitioning, query types, data, and system characteristics. We qualitatively and quantitatively analyze the features of representative partitioning algorithms that impact overall query performance. Along with these analyses, we propose a data sampling-based approach for selecting optimized partitioning strategies. Extensive experiments on large and complex datasets using MapReduce frameworks are conducted to validate the correctness of our analysis and the effectiveness of our optimization approach.

Join order optimization is pivotal in database query optimization, seeking the most efficient join sequence to reduce execution costs. As more tables join, the complexity surges, turning it into an NP-hard problem due to the exponential growth of possible orders. Deep reinforcement learning (DRL) has recently made significant strides, outperforming traditional algorithms by treating join selection as a Markov Decision Process to devise more effective strategies.Current methods struggle with integrating query semantics and plan structures, as well as encountering issues with complex joins where bottom-up learning can lead to information loss.To tackle these issues, we present the Tree-based Selective State Space Models for Efficient Join Order Selection Learning(TESSM). This framework uses the Tree Mamba architecture to integrate join pattern graphs with execution plan nodes, enhancing long-term dependency information flow. A tiered training strategy enhances the model's training precision and speed.Our approach has proven effective, as evidenced by JOB and TPC-H benchmark tests, showing TESSM's substantial improvements in query optimization efficiency and effectiveness.

Predicting anomalies of mobile equipment plays an important role in performing preventive maintenance, alleviating major economic losses and personal safety issues. Previous studies basically adopted data-driven models for anomaly prediction or detection of industrial equipment, ignoring the importance of domain knowledge. The domain knowledge can more accurately and theoretically capture the complex relationship among features. However, building the deep learning models incorporating domain knowledge is very difficult due to the following challenges. First, the domain knowledge is often different from the actual state of the equipment, so it is difficult to obtain knowledge information that conforms to the real situation. Second, domain knowledge is difficult to directly and effectively be applied to deep learning models due to its diverse representations. In this paper, we propose a Multivariate Time-Series Anomaly Prediction with Domain Knowledge (MTAP-DK) to address these issues. Specifically, we firstly propose a knowledge extraction module, which can extract the domain equations that conform to the actual situation with the domain knowledge and historical data. Secondly, we design a domain guidance module to guide and constrain the graph neural network from the knowledge level, to improve its capabilities to express the relationship among features. Thirdly, we predict future data based on the graph incorporating knowledge information. Finally, the prediction is reconstructed by the multi-scale convolution reconstruction method, and the abnormal information is inferred according to the reconstruction error.

Many video websites offer packages composed of paid videos. Users who purchase a package become members of the website, and thus can enjoy the membership service, such as watching the paid videos. It is practically important to predict which users will become members so that the website can recommend them the suitable packages for purchasing. Existing works generally predict the purchase behavior of users through capturing their interests in items. However, such works cannot be directly applied to the studied problem due to the following challenges. First, some important features of videos and packages change over time, such as the number of clicks and the update of the videos. Existing methods are not capable to capture such dynamic features. Second, a user's purchasing intention is very hard to capture. A user watching a video does not necessarily mean that he/she would like to purchase the corresponding package. In this paper, we propose a Dynamic Evolution based Deep Hierarchical Intention Network (Dy-HIEN for short) for membership prediction, which contains two modules. In the first module, we design a dynamic embedding learning method, applying multi-relational heterogeneous information network and attention mechanism to effectively represent the embedding of videos and packages. In the second module, a hierarchical method is proposed to extract the purchase intention of users. First, the video play history is divided into sessions based on the clicks on packages, and then time-order encoder and kernel functions are applied to mine the intention pattern associated with the package clicked in each session. Extensive experiments on real-world datasets are conducted to demonstrate the advantages of the proposed model on a variety of evaluation metrics.

Finding fuel-efficient routes for drivers has increasingly important value in terms of saving energy, protecting the environment and saving expenses. Previous studies basically adopt simple fuel consumption calculation or prediction methods to recommend the fuel-efficient routes within a city, which have two major limitations. First, the effect of drivers' driving behavior preferences (e.g. acceleration, frequency of clutch use, etc.) on fuel consumption is not fully studied and utilized. Second, existing methods mainly focus on short-distance route recommendation. Due to the difference in the road network structure and route composition, it is not effective to directly apply the route recommendation methods designed for short-distance travel within a city on the scenario of long-distance travel among cities. In this paper, we propose a novel model PLd-FeRR for the Personalized Long-distance Fuel-efficient Route Recommendation. Specifically, we first identify the features reflecting the user's driving behavior preference based on the user's historical driving trajectory, and then extract the potential factors that can affect long-distance fuel consumption. As transformer can effectively capture the temporal features for long sequence data, the extracted personalized driving preference features and long-distance fuel consumption features are input into a transformer-based fuel consumption prediction model. Next, the prediction model is combined with a genetic algorithm to further improve the performance of recommending fuel-efficient routes. Extensive evaluations are conducted on the large real-world dataset, and the results show the effectiveness of our proposal.

Traffic flow prediction based on vehicle trajectories collected from the installed GPS devices is critically important to Intelligent Transportation Systems (ITS). One limitation of existing traffic prediction models is that they mostly focus on predicting road-segment level traffic conditions, which can be considered as a fine-grained prediction. In many scenarios, however, a coarse-grained prediction, such as predicting the traffic flows among different urban areas covering multiple road links, is also required to help government have a better understanding on traffic conditions from the macroscopic point of view. This is especially useful in the applications of urban planning and public transportation planning. Another limitation is that the correlations among different types of traffic-related features are largely ignored. For example, the traffic flow and traffic speed are usually negatively correlated. Existing works regard these traffic-related features as independent features without considering their correlations. In this article, we for the first time study the novel problem of multivariate correlation-aware multi-scale traffic flow predicting, and we propose a feature correlation-aware spatio-temporal graph convolutional networks named MC-STGCN to effectively address it. Specifically, given a road graph, we first construct a coarse-grained road graph based on both the topology closeness and the traffic flow similarity among the nodes (road links). Then a cross-scale spatial-temporal feature learning and fusion technique is proposed for dealing with both the fine- and coarse-grained traffic data. In the spatial domain, a cross-scale GCN is proposed to learn the multi-scale spatial features jointly and fuse them together. In the temporal domain, a cross-scale temporal network that is composed of a hierarchical attention is designed for effectively capturing intra- and inter-scale temporal correlations. To effectively capture the feature correlations, a feature correlation learning component is also designed. Finally, a structural constraint is introduced to make the predictions on the two scale traffic data consistent. We conduct extensive evaluations over two real traffic datasets, and the results demonstrate the superior performance of the proposal on both fine- and coarse-grained traffic predictions.

Fault prediction is critically important for many mobile equipments such as vehicles, ships and spacecrafts. Sensors deployed on these equipments continuously collect the status data, which are usually multivariate time series data. It is challenging to accurately predict the failure of the equipments based on the generated multivairate time series due to the complex correlations among the variables and the dynamic operation conditions. Though many methods have been proposed, they are not effective to provide an interpretable and accurate fault prediction result. This paper proposes a two-stage Interpretable Fault Prediction method based on Anomaly Detection and Anomaly Accumulation, called IFP-ADAC. Specially, we first design an anomaly detection module based on Generative Adversarial Nets due to the lack of samples. The generator captures the correlations among multiple variables and the temporal dependency within each variable jointly. Second, we design an anomaly accumulation model based on LSTM to capture the anomaly growth pattern, and the attention mechanism has been introduced to consider the severity of the detected anomalies. Compared with the end- to-end methods, our two-stage fault prediction method based on anomaly detection and accumulation has better interpretability. Extensive experiments conducted on two real-world datasets show the superior performance of our method.

Fuel-efficient route recommendation has been increasingly valuable for both energy conservation and environmental protection. Most existing methods analyze fuel consumption factors from the short-distance trajectories. However, due to the differences in the road network structure and route composition between long-distance and short-distance trajectories, directly using these methods to recommend fuel-efficient routes for long-distance drivers is less effective. In addition, previous works usually adopt heuristic algorithms due to efficiency, but empirically set the heuristic functions, which makes it difficult to integrate various influencing factors appropriately. In this paper, we propose a novel fuel-efficient route recommendation model for long-distance driving, LDFeRR. We first identify the potential factors that affect fuel consumption over long distances based on historical trajectories. To fully exploit these factors to estimate fuel consumption and further provide reliable recommendations, we propose to integrate deep learning methods with heuristic algorithm. Specially, we use a multi-layer perceptron (MLP) to predict the fuel consumption of a single road segment. We also employ an attention-based bidirectional gated recurrent unit (Att-BiGRU) to estimate the fuel consumption between two locations connected by routes. In this way, the two cost functions of the classic heuristic algorithm, i.e., A* algorithm, are automatically learned. Extensive experiments on the large real-world dataset demonstrate the effectiveness of our proposed model.

As one promising way to solve the challenging issues of data sparsity and cold start in recommender systems, crossdomain recommendation has gained increasing research interest recently. Cross-domain recommendation aims to improve the recommendation performance by means of transferring explicit or implicit feedback from the auxiliary domain to the target domain. Although the side information of review texts and item contents has been proven to be useful in recommendation, most existing works only use one kind of side information and cannot deeply fuse this side information with ratings. In this paper, we propose a Review and Content based Deep Fusion Model named RC-DFM for crossdomain recommendation. We first extend Stacked Denoising Autoencoders (SDAE) to effectively fuse review texts and item contents with the rating matrix in both auxiliary and target domains. Through this way, the learned latent factors of users and items in both domains preserve more semantic information for recommendation. Then we utilize a multi-layer perceptron to transfer user latent factors between the two domains to address the data sparsity and cold start issues. Experimental results on real datasets demonstrate the superior performance of RC-DFM compared with state-of-the-art recommendation methods.Deeply Fusing Reviews and Contents for Cold Start Users in Cross-Domain Recommendation Systems

Relevance measure in both homogeneous and heterogeneous networks has been extensively studied. However, how to measure the relevance among different-typed objects in weighted signed heterogeneous information networks remains an open problem. It is challenging to incorporate both positive and negative multi-typed relationships simultaneously in signed heterogeneous networks due to the opposite opinions implied by them. To this end, this paper proposes a random walk based approach for relevance measure by utilizing and modeling the rich semantic information in weighted signed heterogeneous networks. Particularly, we first transform a signed network into a non-signed network according to the different semantic meanings represented by positive and negative relationships. This paves the way to properly utilize negative relationships. Next, we conduct random walk from the source object to the target object based on a bunch of single meta-paths separately. Finally, we combine multiple meta-paths together to obtain a more comprehensive relatedness between the source object and the target object. Extensive experiments on real datasets demonstrate the superior performance of the proposed approach.

How to model the influence propagation accurately in social network is a critical and challenge task. Although numerous attempts have been made for this topic, few of them consider the user’s negative influence. Positive influence will encourage people to perform some action while the negative one will degrade the probability. Thus, it is meaningful to model the influence propagation by considering both the positive and negative influence. What’s more, previous research is mostly based on the assumption that the influence probabilities between users are known, however, they are typically unknown in real-world social networks. To address these problems, a novel Multipolar Factors aware Independent Cascade model (MFIC) is proposed to outline the information diffusion in social network. Then, the user-to-user influence probability is learnt with the users’ behavior logs based on the EM algorithm. We also apply the discovered influence probabilities to user behavior prediction. Experiments are conducted over real data sets, Flixster and Digg, validating the effectiveness of our methods.

Relevance search in heterogeneous information networks is a basic and crucial operation which is usually used in recommendation, clustering and anomaly detection. Nowadays most existing relevance search methods focus on objects in homogeneous information networks. In this paper, we propose a method to find the top-k most relevant objects to a specific one in heterogeneous networks. It is a two phase process that we get the initial relevance score based on the method of pair wise random walk along given meta-paths, which is a meta-level description of the path instances in heterogeneous information networks, and then take user preference into consideration to calculate the weights combination of meta-paths and model the problem into a multi-objective linear planning problem which can be solved with the method of generic algorithm. Besides, to ensure the efficiency, we use matrix computation and selective materialization to avoid the recursive computation of pair wise random walk. What's more, we propose an effective pruning method to skip unnecessary objects computations. The experiments on IMDB and DBLP dataset show that the method can gain a better accuracy and efficiency.

The problem of prediction is an important task in network analysis, which has attracted more attention from computer science communities. In this paper, prediction in signed heterogeneous networks is addressed, which contains two aspects, link prediction and sign prediction. Most of previous studies focus on non-signed networks that have only positive links or homogeneous networks that have only one type of nodes. However, there are many signed heterogeneous networks in which the nodes and links belong to different types and links can be either positive (indicating relationships such as trust, preferences, friendship, and etc) or negative (indicating relationships such as distrust, dislike, opposition, and etc) in real world. For link prediction, a rule-based methodology called RulePredict is proposed in the paper. In RulePredict, we first extract all features systematically which contain positive features that promote the existence of links and negative ones that reduce the possibility reversely. Then, the weights associated with different features will be learned by a supervised method based on generalized least squares (GLS). For sign prediction, we put forward a new method called HeteSign to calculate the polarity of the links based on the similarity of two objects depends on their linked objects in heterogeneous networks. Experiments are conducted on real networks, the IMDB and Epinions networks, which demonstrate that our approach gets better performance in terms of accuracy.

A modified collaborative filtering recommendation has been put forward for clustered data in social networks. Firstly, the basic model is built up, which shows the outline of current existing types of the recommendation platform. Secondly, on this basis, the vertexes and edges in the model are divided as the entities of the social networks with properties given formally. This framework implements traditional collaborative filtering based on clustered datasets using improved k-means clustering method, and our design has put forward to picking up the active users among all of them and let them help our system generate recommendations more precise and faster. Finally, we have crawled through data from two famous movie recommendation websites, MovieLens and Imdb, in the experiment test. From the experiment results, it is concluded that the implemented collaborative filtering recommendation system is performed better than the naive one both in precision and stability, which aims to supply collaborative filtering recommendation and actively contributing recommendation that suits users' tastes, has completed the task of improving collaborative recommendation and reached the expected goals.

As an important supporting technology of Market Intelligence (MI), Web data integration is facing new challenges, such as the integrity of data acquisition, the quality of data extraction and data consolidation. To solve such problems, we propose an MI-oriented web data integration system (MI-WDIS), which achieves excellent performances in integrating Surface Web and Deep Web data with much less manual work. Based on MI-WDIS, we have developed a platform for intelligent analysis of job data. The platform collects tens of thousands of job data daily and provides personalized services for job seekers through diversified channels. Besides, it provides other advanced services, including intelligence analysis, automatic monitoring and alerting, for various organizations, such as enterprises, training institutions and recruitment agencies.