With the fast development of various positioning techniques such as Global Position System (GPS), mobile devices and remote sensing, spatio-temporal data has become increasingly available nowadays. ...Mining valuable knowledge from spatio-temporal data is critically important to many real-world applications including human mobility understanding, smart transportation, urban planning, public safety, health care and environmental management. As the number, volume and resolution of spatio-temporal data increase rapidly, traditional data mining methods, especially statistics based methods for dealing with such data are becoming overwhelmed. Recently deep learning models such as recurrent neural network (RNN) and convolutional neural network (CNN) have achieved remarkable success in many domains, and are also widely applied in various spatio-temporal data mining (STDM) tasks such as predictive learning, anomaly detection and classification. In this paper, we provide a comprehensive review of recent progress in applying deep learning techniques for STDM. We first categorize the spatio-temporal data into five different types, and then briefly introduce the deep learning models that are widely used in STDM. Next, we classify existing literature based on the types of spatio-temporal data, the data mining tasks, and the deep learning models, followed by the applications of deep learning for STDM in different domains.
•Extract both the mean and trend features based on Symbolic Aggregate approXimation.•Design single classifier based on both the mean and trend features.•Construct ensemble classifier by multi-feature ...dictionary and ensemble learning.•Experiments on real datasets verify effectiveness of our proposal.
Time series classification is an important task for mining time series data, and many high level representations of time series have been proposed to address it. Symbolic Aggregate approXimation (SAX) is a classic high level symbolic representation method which can effectively reduce the dimensionality of time series. However, SAX-based methods for time series classification cannot achieve promising results, because SAX only extracts the mean feature of subsequence to make symbolization. In this paper, we present a novel ensemble method based on SAX called TBOPE, which is based on multi-feature dictionary representation and ensemble learning. Specifically, we first extract both the mean feature and trend feature of time series. Second, we create the histograms of two kinds of feature based on the Bag-of-Feature mode and construct multiple single classifiers. Finally, we build an ensemble classifier to improve the classification performance. Experimental results on various time series datasets have shown that the proposed method is competitive to state-of-the-art methods.
In recent years, early classification on time series has become increasingly important in time-sensitive applications. Existing shapelet based methods still cannot work well on this problem. First, ...the effectiveness of traditional shapelet based methods would be influenced by the number of shapelet candidates. Second, it is difficult for previous methods to obtain diverse shapelets in shapelet selection. In this paper, we propose an Improved Early Distinctive Shapelet Classification method named IEDSC. We first present a new method to more precisely measure the similarity between time series, which takes into account of the relative trend of time series. Second, in shapelet extraction, we propose a pruning technique to reduce the number of shapelets by predicting the starting positions of shapelets with good quality. In addition, a new shapelet selection method is also proposed to remove the similar shapelets, so as to maintain the diversity of shapelets. Finally, the experimental results on 16 benchmark datasets show that the proposed method outperforms state-of-the-art for early classification on time series.
With the advancement of location acquisition technologies, a large amount of raw global positioning system (GPS) trajectory data is produced by many moving devices. Learning transportation modes from ...the GPS trajectory data is an important problem in the domain of trajectory data mining. Traditional supervised learning‐based approaches rely heavily on data preprocessing and feature engineering, which require domain expertise and are time consuming. The authors propose a deep learning‐based convolutional long short term memory (LSTM) model for transportation mode learning, in which the convolution neural network is first used to extract deep high‐level features and then LSTM is used to learn the sequential patterns in the data that uses both GPS and weather features, thus making the full use of spatiotemporal operations. The authors have also analysed the impact of the geospatial region on human mobility. Experiments conducted on the Microsoft Geolife data set fused with the weather data set show that their model achieves the state‐of‐the‐art results. The authors compare the performance of their model with the benchmark models, which shows the superiority of their model having 3% improvement in accuracy using only GPS features, and the accuracy is further improved by 4 and 7% on including the impact of geospatial region and weather attributes, respectively.
Network embedding aims to learn a low-dimensional representation vector for each node while preserving the inherent structural properties of the network, which could benefit various downstream mining ...tasks such as link prediction and node classification. Most existing works can be considered as generative models that approximate the underlying node connectivity distribution in the network, or as discriminate models that predict edge existence under a specific discriminative task. Although several recent works try to unify the two types of models with adversarial learning to improve the performance, they only consider the local pairwise connectivity between nodes. Higher-order structural information such as communities, which essentially reflects the global topology structure of the network, is largely ignored. To this end, we propose a novel framework called
CANE
to simultaneously learn the node representations and identify the network communities. The two tasks are integrated and mutually reinforce each other under a novel adversarial learning framework. Specifically, with the detected communities,
CANE
jointly minimizes the pairwise connectivity loss and the community assignment error to improve node representation learning. In turn, the learned node representations provide high-quality features to facilitate community detection. Experimental results on multiple real datasets demonstrate that
CANE
achieves substantial performance gains over state-of-the-art baselines in various applications including link prediction, node classification, recommendation, network visualization, and community detection.
As one of the most influential social media platforms, microblogging is becoming increasingly popular in the last decades. Each day a large amount of events appear and spread in microblogging. The ...spreading of events and corresponding comments on them can greatly influence the public opinion. It is practical important to discover new emerging events in microblogging and predict their future popularity. Traditional event detection and information diffusion models cannot effectively handle our studied problem, because most existing methods focus only on event detection but ignore to predict their future trend. In this paper, we propose a new approach to detect burst novel events and predict their future popularity simultaneously. Specifically, we first detect events from online microblogging stream by utilizing multiple types of information, i.e., term frequency, and user׳s social relation. Meanwhile, the popularity of detected event is predicted through a proposed diffusion model which takes both the content and user information of the event into account. Extensive evaluations on two real-world datasets demonstrate the effectiveness of our approach on both event detection and their popularity prediction.
GPS datasets in the big data regime provide rich contextual information that enable efficient implementation of advanced features such as navigation, tracking, and security in urban computing ...systems. Understanding the hidden patterns in large amount of GPS data is critically important in ubiquitous computing. The quality of GPS data is the fundamental key problem to produce high quality results. In real world applications, certain GPS trajectories are sparse and incomplete; this increases the complexity of inference algorithms. Few of existing studies have tried to address this problem using complicated algorithms that are based on conventional heuristics; this requires extensive domain knowledge of underlying applications. Our contribution in this paper are two-fold. First, we proposed deep learning based bidirectional convolutional recurrent encoder-decoder architecture to generate the missing points of GPS trajectories over occupancy grid-map. Second, we interfaced attention mechanism between enconder and decoder, that further enhance the performance of our model. We have performed the experiments on widely used Microsoft geolife trajectory dataset, and perform the experiments over multiple level of grid resolutions and multiple lengths of missing GPS segments. Our proposed model achieved better results in terms of average displacement error as compared to the state-of-the-art benchmark methods.
Origin-Destination (OD) prediction which aims to predict the number of passenger’s travel demands from one region to another, is critically important to many real applications including intelligent ...transportation systems and public safety. The challenges of this problem lie in both the dynamic patterns of the human mobility data and data sparsity in issue in some regions. Thus it is difficult to model the complex spatio-temporal correlations of the human mobility data to predict the OD of their trips. Meanwhile, the crowd flows in different regions of a city and the context features (e.g. holiday, weather and POIs) are potentially useful to alleviate the data sparsity issue and improve the OD prediction, but are largely ignored by existing works. In this paper, we propose a deep spatio-temporal framework which named Auxiliary-tasks Enhanced Spatio-Temporal Network (AEST) to more effectively address the OD prediction problem. AEST trains a model to conduct OD inference via learning crowd flow and external data as auxiliary task. The novel Hierarchical Convolutional LSTM (HC-LSTM) Network is proposed which combines CNN, GCN and LSTM to effectively capture spatiao-temporal correlations. In addition, we design a Contextual Network (ContextNet) which learns representations of contextual information to assist OD prediction. We conduct extensive experiments over bike and taxicab trip datasets in New York. The results show that our method is superior to the state-of-art approaches.
Complex networks are everywhere, such as the power grid network, the airline network, the protein-protein interaction network, and the road network. The networks are 'robust yet fragile', which means ...that the networks are robust against random failures but fragile under malicious attacks. The cascading failures, system-wide disasters and intentional attacks on these networks are deserving of in-depth study. Researchers have proposed many solutions to improve the robustness of these networks. However whilst many solutions preserve the degree distribution of the networks, little attention is paid to the community structure of these networks. We argue that the community structure of a network is a defining characteristic of a network which identifies its functionality and thus should be preserved. In this paper, we discuss the relationship between robustness and the community structure. Then we propose a 3-step strategy to improve the robustness of a network, while retaining its community structure, and also its degree distribution. With extensive experimentation on representative real-world networks, we demonstrate that our method is effective and can greatly improve the robustness of networks, while preserving community structure and degree distribution. Finally, we give a description of a robust network, which is useful not only for improving robustness, but also for designing robust networks and integrating networks.
Collaborative filtering (CF) is a widely adopted technique in recommender systems. Traditional CF models mainly focus on predicting the user preference to items in a single domain, such as the movie ...domain or the music domain. A major challenge for such models is the data sparsity, and especially, CF cannot make accurate predictions for the cold-start users who have no ratings at all. Although cross-domain collaborative filtering (CDCF) is proposed for effectively transferring knowledge across different domains, it is still difficult for existing CDCF models to tackle the cold-start users in the target domain due to the extreme data sparsity. In this paper, we propose the cross-domain latent feature mapping (CDLFM) model for the cold-start users in the target domain. Firstly, in order to alleviate the data sparsity in single domain and provide essential knowledge for next step, we take users’ rating behaviors into consideration and propose the matrix factorization by incorporating user similarities. Next, to transfer knowledge across domains, we propose the neighborhood-based cross-domain latent feature mapping method. For each cold-start user, we learn his/her feature mapping function based on his/her neighbor linked users. By adopting gradient boosting trees and multilayer perceptron to model the cross-domain feature mapping function, two CDLFM models named CDLFM-GBT and CDLFM-MLP are proposed. Experimental results on two real datasets demonstrate the superiority of our proposed model against other state-of-the-art methods.