Search

Filter pruning is an essential method for compressing and accelerating deep neural networks on computationally restricted devices. Despite recognizing the high correlation between filter redundancy and samples, existing methods primarily focus on independently searching for optimal subnetworks from individual input while ignoring the relationships among different inputs. In this paper, we propose a novel approach called Gated Filter Pruning based on Sample Manifold Relationships, which exploits and aligns the manifold relationships of all samples during training to obtain an optimal subnetwork. Firstly, we introduce a Gated Filter Normalization Module (GFNM) that excavates the manifold information of each sample, applicable to the operator level without adding many additional parameters. GFNM incorporates explainable control variables jointly optimized with convolutional weights, explicitly determining the competition and cooperation among filters during training. Subsequently, Manifold Regularized Pruning Module (MRPM) measures the manifold relationships between samples and subnetworks, efficiently regularizing the solution space of sample-network pairs. The manifold relationships between samples and subnetworks are aligned in training to derive an effective subnetwork for all input samples. Extensive experimental results validate the effectiveness of our method, demonstrating competitive performance in terms of accuracy and computational cost compared to state-of-the-art (SOTA) methods. [ABSTRACT FROM AUTHOR]

Filter pruning is an essential method for compressing and accelerating deep neural networks on computationally restricted devices. Despite recognizing the high correlation between filter redundancy and samples, existing methods primarily focus on independently searching for optimal subnetworks from individual input while ignoring the relationships among different inputs. In this paper, we propose a novel approach called Gated Filter Pruning based on Sample Manifold Relationships, which exploits and aligns the manifold relationships of all samples during training to obtain an optimal subnetwork. Firstly, we introduce a Gated Filter Normalization Module (GFNM) that excavates the manifold information of each sample, applicable to the operator level without adding many additional parameters. GFNM incorporates explainable control variables jointly optimized with convolutional weights, explicitly determining the competition and cooperation among filters during training. Subsequently, Manifold Regularized Pruning Module (MRPM) measures the manifold relationships between samples and subnetworks, efficiently regularizing the solution space of sample-network pairs. The manifold relationships between samples and subnetworks are aligned in training to derive an effective subnetwork for all input samples. Extensive experimental results validate the effectiveness of our method, demonstrating competitive performance in terms of accuracy and computational cost compared to state-of-the-art (SOTA) methods. [ABSTRACT FROM AUTHOR]

Industrial image anomaly detection achieves automated detection and localization of defects or abnormal regions in images through image processing and deep learning techniques. Currently, utilizing the approach of reverse knowledge distillation has yielded favourable outcomes. However, it is still a challenge in terms of the feature extraction capability of the image and the robustness of the decoding of the student network. This study first addresses the issue that the teacher network has not been able to extract global information more effectively. To acquire more global information, a vision transformer network is introduced to enhance the model's global information extraction capability, obtaining better features to further assist the student network in decoding. Second, for anomalous samples, to address the low similarity between features extracted by the teacher network and features restored by the student network, Gaussian noise is introduced. This further increases the probability that the features decoded by the student model match normal sample features, enhancing the robustness of the student model. Extensive experiments were conducted on industrial image datasets AeBAD, MvtecAD, and BTAD. In the AeBAD dataset, under the PRO performance metric, the result is 89.83%, achieving state‐of‐the‐art performance. Under the AUROC performance metric, it reaches 83.35%. Similarly, good results were achieved on the MvtecAD and BTAD datasets. The proposed method's effectiveness and performance advantages were validated across multiple industrial datasets, providing a valuable reference for the application of industrial image anomaly detection methods. [ABSTRACT FROM AUTHOR]

Knowledge distillation is a common and effective method in model compression, which trains a compact student model to mimic the capability of a large teacher model to get superior generalization. Previous works on knowledge distillation are underperforming for challenging tasks such as object detection, compared to the general application of unsophisticated classification tasks. In this paper, we propose that the failure of knowledge distillation on object detection is mainly caused by the imbalance between features of informative and invalid background. Not all background noise is redundant, and the valuable background noise after screening contains relations between foreground and background. Therefore, we propose a novel regional filtering distillation (RFD) algorithm to solve this problem through two modules: region selection and attention-guided distillation. Region selection first filters massive invalid backgrounds and retains knowledge-dense regions on near object anchor locations. Attention-guided distillation further improves distillation performance on object detection tasks by extracting the relations between foreground and background to migrate key features. Extensive experiments on both one-stage and two-stage detectors have been conducted to prove the effectiveness of RFD. For example, RFD improves 2.8% and 2.6% mAP for ResNet50-RetinaNet and ResNet50-FPN student networks on the MS COCO dataset, respectively. We also evaluate our method with the Faster R-CNN model on Pascal VOC and KITTI benchmark, which obtain 1.52% and 4.36% mAP promotions for the ResNet18-FPN student network, respectively. Furthermore, our method increases 5.70% of mAP for MobileNetv2-SSD compared to the original model. The proposed RFD technique performs highly on detection tasks through regional filtering distillation. In the future, we plan to extend it to more challenging task scenarios, such as segmentation. [ABSTRACT FROM AUTHOR]

Semi-supervised learning uses a small amount of labeled data to guide the model and a large amount of unlabeled data to improve its performance. Most semi-supervised object detection methods build a teacher-student architecture and train the student network with pseudo-labels generated by the teacher. To guarantee the accuracy of pseudo-labels, a high threshold value is always applied to filter out all of the low-scoring labels according to the inference results of the teacher. We argue that these discarded labels with low scores have more information for the current model and should be taken into consideration in the training phase. Further, we propose an uncertain region mining(URM) framework that utilizes these uncertainty low confidence labels. Especially, URM exploits the uncertain labels from two aspects: (1)Recalling the underlying correct labels: URM designs a fusion function that rectifies the outputs of the teacher with the student and recalls the high-quality pseudo-labels. (2)Avoiding the error information of the error labels: URM proposes a negative loss function that utilizes the uncertain labels without introducing error information. For the regression task, a new branch is attached to the detector to predict the localization scores of bounding boxes. Based on the predicted scores, we propose a re-weighting strategy that alleviates the noisy problem from the imprecise localization of the bounding boxes. Experiments on PASCAL-VOC, MS-COCO and DOTA datasets demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Overparameterization of convolutional neural networks allows model compression, and model pruning algorithms have attracted much interest due to their practical acceleration effects. The pruning algorithm should remove as many redundant structures as possible while maintaining the original accuracy to maximize the acceleration effect. However, prior works utilize benign samples to evaluate the loss of accuracy but ignore the vulnerability to malicious adversarial samples, which brings potential security risks. To tackle this limitation, we propose the distortion diminishing with vulnerability filters pruning (DD-VFP) method that simultaneously improves the compressed model’s classification accuracy and adversarial defense capabilities. Specifically, we define adversarial vulnerability for filters by computing the distortion of their latent features. We then propose a new pruning framework, vulnerability filter pruning (VFP), to remove filters with higher adversarial vulnerability during training. We propose a new loss for adversarial training, distortion diminishing (DD) loss, which directly suppresses the model’s dependence on non-robust features by reducing the latent feature distortion under adversarial perturbation. Experiments on multiple benchmark datasets prove the effectiveness of our method, which not only achieves state-of-the-art adversarial defense capabilities but also removes more model parameters. An interesting phenomenon is that although the DD-VFP method slightly loses accuracy after pruning, the trade-off between accuracy and adversarial defense capabilities is significantly reduced, which proves that our method successfully removes non-robust features. [ABSTRACT FROM AUTHOR]

We propose GADA-SegNet, a gated attentive domain adaptation network for semantic segmentation of LiDAR point clouds. Unlike most of existing methods that learn fully from point-wise annotations, our GADA-SegNet attempts to learn from labeled data first and then transfer itself smoothly to unlabeled data. We have three key contributions to bridge the domain gap between the labeled data and the unlabeled yet unseen data. First, we design a new gated connection module that can filter out noise and domain-private features from the low-level features, for better high- and low-level feature fusion. Second, we introduce a multi-scale attention module that can ease the large-scale variation of objects and class imbalance in complex scenes to reduce the class-level domain gap. Third, we develop a shared domain discriminator to implement the class-level domain discrimination for large-scale LiDAR point clouds. Experiments on both synthetic-to-real and real-to-real scenarios show clear improvements of our GADA-SegNet over its competitors. [ABSTRACT FROM AUTHOR]

Text image recognition in natural scenes is challenging in computer vision, even though it is already widely used in real-life applications. With the development of deep learning, the accuracy of scene text recognition has been continuously improved. The encoder-decoder architecture is currently a general framework in scene text recognition. With 2D attention on the decoder, better attention can be paid to the position of each character. However, many methods based on the encoder-decoder architecture only adopt the attention mechanism on the decoder. Therefore, their ability to locate characters is limited. In order to solve this problem, we propose a Transformer-based encoder-decoder structure with a two-stage attention mechanism for scene text recognition. At the encoder, a first-stage attention module integrating spatial attention and channel attention is used to capture the overall location of the text in the image, while at the decoder, a second-stage attention module is used to pinpoint the position of each character in the text image. This two-stage attention mechanism can locate the position of the text more effectively and improve recognition accuracy. Also, we design a multi-branch feature fusion module for the encoder that can fuse features from different receptive fields to obtain more robust features. We train the model on synthetic text datasets and test it on real scene text datasets. The experimental results show that our model is very competitive. [ABSTRACT FROM AUTHOR]

Industrial defect detection plays an important role in smart manufacturing and is widely used in various scenarios such as smart inspection and product quality control. Currently, although utilizing a framework for knowledge distillation to identify industrial defects has achieved great progress, it is still a significant challenge task to extract better image features and prevent overfitting for student networks. In this study, a reverse knowledge distillation framework with two teachers is designed. First, for the teacher network, two teachers with different architectures are used to extract the diverse features of the images from multiple models. Second, considering the different contributions of channels and different teacher networks, the attention mechanism and iterative attention feature fusion idea are introduced. Finally, to prevent overfitting, the student network is designed with a network architecture that is inconsistent with the teacher network. Extensive experiments were conducted on Mvtec and BTAD datasets, which are industrial defect detection datasets. On the Mvtec dataset, the average accuracy values of image-level and pixel-level ROC achieved 99.43% and 97.87%, respectively. On the BTAD dataset, the average accuracy values of image-level and pixel-level ROC reached 94% and 98%, respectively. The performance on both datasets is significantly improved, demonstrating the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Due to the high cost of mobile nodes and the low cost of fixed nodes, in some applications of wireless sensor networks, in order to reduce the use cost, the mixed mode is used for node deployment. The deployed nodes include both fixed and mobile nodes. After the nodes are deployed, the fixed nodes do not move, and the mobile nodes move reasonably according to the network coverage to improve the network coverage effect. Due to the differences in equipment structure and use cost between fixed nodes and mobile nodes, in order to complete the monitoring task under the condition of reducing the use cost of wireless sensor networks, the number and position of mobile nodes can be reasonably adjusted as needed to improve the effective coverage area of wireless sensor networks. In the process of adjusting the position of mobile nodes, the moving distance of mobile nodes is fully considered. Simulation results show that this algorithm can improve the coverage effect of mixed wireless sensor networks and reduce the moving distance of mobile nodes. [ABSTRACT FROM AUTHOR]

Extraction and analysis of the EEG (electroencephalograph) information features generated during behavioral decision-making can provide a better understanding of the state of mind. Previous studies have focused more on the brainwave features after behavioral decision-making. In fact, the EEG before decision-making is more worthy of our attention. In this study, we introduce a new index based on the reaction time of subjects before decision-making, called the Prestimulus Time (PT), which have important reference value for the study of cognitive function, neurological diseases, and other fields. In our experiments, we use a wearable EEG feature signal acquisition device and a systematic reward and punishment experiment to obtain the EEG features before and after behavioral decision-making. The experimental results show that the EEG generated after behavioral decision due to loss is more intense than that generated by gain in the medial frontal cortex (MFC). In addition, different characteristics of EEG signals are generated prior to behavioral decisions because people have different expectations of the outcome. It will produce more significant negative-polarity event-related potential (ERP) in the forebrain area when the humans are optimistic about the outcomes. [ABSTRACT FROM AUTHOR]

High-quality network coverage can quickly and accurately collect the data in the monitoring area and complete the task of target monitoring in wireless sensor networks. The existing related work focuses on improving the effect of network coverage and reducing the consumption of network resources. Aiming at the problems of coverage blind area and node redundancy in network coverage, a monitoring area coverage algorithm based on an improved virtual algorithm is proposed by improving the existing virtual force algorithm. This algorithm controls the moving direction and distance of nodes by adjusting the distance between nodes. Make the node move to a reasonable position, improve the coverage effect of the network, and reduce the moving distance of the node. The simulation results show that compared with the DNDVF algorithm and VFADP algorithm, this algorithm has a better application effect, can improve the coverage effect, and reduce the moving distance of nodes. [ABSTRACT FROM AUTHOR]

Wireless sensor networks deploy a large number of wireless sensor nodes in the monitoring area to complete the target monitoring task. The coverage effect is an important index to evaluate wireless sensor networks. The effect of network coverage is realized by the wireless sensor nodes deployed in the monitoring area, because the reasonable deployment of sensor nodes can affect the accuracy, reliability, real time, and integrity of monitoring target data collection. The main purpose of network coverage is to use the least number of wireless sensor nodes to complete the monitoring task of targets in the monitoring area and to avoid the emergence of coverage blind areas. Random node deployment, electronic equipment failure, software error, fire spread, flood, and other phenomena may lead to large-scale failure of wireless sensor nodes. For the failure of some nodes, the phenomenon of coverage blind area appears in the network. In order to repair the coverage blind area in the network, a coverage blind area repair algorithm for wireless sensor networks is proposed in this article. The coverage blind area attracts the nodes, makes the nodes move, and repairs the coverage blind area in the network. [ABSTRACT FROM AUTHOR]

Summary: Coverage is one of the basic problems that any wireless sensor network application must solve properly to achieve the expected quality of service. The communication distance of wireless sensor network nodes is limited, the initial energy is low, and many optimization objectives need to be considered in the deployment, which makes the network deployment optimization very complicated, and many problems are difficult. In essence, the deployment problem is a multidimensional nonlinear optimization problem with continuous discrete variables. In order to meet the needs of network monitoring, this article establishes a reasonable mathematical model according to the research objectives. By adjusting node position to obtain better network coverage effect, an optimization algorithm of monitoring area coverage based on control node position is proposed. By calculating the new location of nodes, more monitoring areas can be covered by the nodes. The simulation results show that the proposed algorithm can improve the coverage effect rapidly. [ABSTRACT FROM AUTHOR]

The semantic representation of deep features is essential for image context understanding, and effective fusion of features with different semantic representations can significantly improve the model's performance on salient object detection. This paper proposes a novel method called multi‐receptive and parallel integration, for salient object detection. Firstly, a multi‐receptive enhancement module is designed to effectively expand the receptive fields of features from different layers and generate features with different receptive fields. Multi‐receptive enhancement module can enhance the semantic representation and improve the model's perception of the image context, which enables the model to locate the salient object accurately. Secondly, in order to reduce the reuse of redundant information in the complex top‐down fusion method and weaken the differences between semantic features, a relatively simple but effective parallel fusion strategy is proposed. It allows multi‐scale features to better interact with each other, thus improving the overall performance of the model. Experimental results on multiple datasets demonstrate that the proposed method outperforms state‐of‐the‐art methods under different evaluation metrics. [ABSTRACT FROM AUTHOR]

Weber's law reveals the relationship between human perception and perceptual stimuli. Inspired by the theory, this paper designs a multi-level convolution correlation feature statistic method for image retrieval. Firstly, the difference between a central pixel and its neighbors is described by Weber's law through computing the differential excitation of image. Then, a multi-level saliency map is obtained by binary transformation and convolution operation. Thirdly, to exploit spatial correlation information of the image, a pixels pair-wise correlation and hierarchy statistic model is constructed. Finally, all intermediate features are concatenated into one histogram, which includes salient color and texture features. Extensive experiments demonstrate the proposed method of this paper has excellent performance. [ABSTRACT FROM AUTHOR]

Vehicle detection in aerial images is an interesting and challenging task. Traditional methods are based on sliding‐window search and handcrafted features, which limits the representation power and has heavy computational costs. Recent research have shown that deep‐learning algorithms are widely used in the field of object detection. However, the deep‐learning algorithms still face many difficulties and challenges in the object detection under the aerial scene. Meanwhile, the high computational cost of detection algorithms lead to low‐detection efficiency. In this study, we build a fast and accurate lightweight detection framework for vehicle detection in aerial scenes. The proposed detection method improves the expressive ability of detection network and significantly reduces the amount of calculations in the model. Meanwhile, setting suitable anchor boxes according to the size of the object vehicles have been introduced in our model, which also effectively improves the performance of the detection. In addition, we have published a new aerial vehicle image dataset and verified the effectiveness of our method. In the Munich dataset and our dataset, our method achieves 85.8% and 91.2% of the mean average precision (mAP), and its detection time is 1.78 and 0.048 s on Nvidia Titan XP. Our results show that the proposed framework achieves significant improvement over several alternatives and state‐of‐the‐art schemes with higher accuracy and less detection time. [ABSTRACT FROM AUTHOR]

Wireless sensor networks have been widely researched and developed in recent years. The node deployment problem is a multi-dimensional nonlinear optimization problem with continuous discrete variables. In order to improve the coverage effect of wireless sensor networks, a network coverage algorithm based on evidence theory is proposed. The motion direction of the wireless sensor node is calculated. The wireless sensor node is moved to the area with low perception probability. Improve the network coverage effect. Reduce the moving distance of wireless sensor nodes. Extended the service time of wireless sensor network. The simulation results show that this algorithm can improve the coverage effect of the monitoring area and reduce the moving distance of nodes effectively. [ABSTRACT FROM AUTHOR]

With the rapid development of the electronic industry, the defect detection of printed circuit board (PCB) components is becoming more and more important. The types of PCB components are diverse and accompanied by complex character information, which is difficult to identify. The traditional detection method is inefficient, and it is unable to effectively perform the diversified category detection of PCB components and character recognition in complex scenes. The deep convolutional neural network has obvious advantages in object detection and character recognition, which can be used to implement a PCB component defect detection system. In this study, the authors have established a lightweight PCB type detection model called LD‐PCB, which can perform real‐time detection while improving detection accuracy. In addition, in the character detection of PCB, they have established a fast and robust character recognition model, called CR‐PCB. This model can effectively improve the accuracy of irregular character recognition. Finally, they established and published a dataset of PCB components, and combined with LD‐PCB and CR‐PCB to realise the PCB defect detection system. This system can realise the functions of defect detection, wrong insertion, missing insertion, and character recognition in industrial PCB production. The results show that the method proposed in this study can effectively detect defects on PCB components. [ABSTRACT FROM AUTHOR]

With the rapid development of the electronic industry, the defect detection of printed circuit board (PCB) components is becoming more and more important. The types of PCB components are diverse and accompanied by complex character information, which is difficult to identify. The traditional detection method is inefficient, and it is unable to effectively perform the diversified category detection of PCB components and character recognition in complex scenes. The deep convolutional neural network has obvious advantages in object detection and character recognition, which can be used to implement a PCB component defect detection system. In this study, the authors have established a lightweight PCB type detection model called LD-PCB, which can perform real-time detection while improving detection accuracy. In addition, in the character detection of PCB, they have established a fast and robust character recognition model, called CR-PCB. This model can effectively improve the accuracy of irregular character recognition. Finally, they established and published a dataset of PCB components, and combined with LD-PCB and CR-PCB to realise the PCB defect detection system. This system can realise the functions of defect detection, wrong insertion, missing insertion, and character recognition in industrial PCB production. The results show that the method proposed in this study can effectively detect defects on PCB components. [ABSTRACT FROM AUTHOR]

Summary: Nowadays, numerous incentive mechanisms of mobile crowdsensing have been designed to attract extensive user participation, but most of these mechanisms focus only on independent task scenarios, where the sensing tasks are independent of each other. On the contrary, we focus on a periodical task scenario, where each user participates in the same type of sensing tasks periodically. In this paper, we consider the long‐term user participation incentive in a general periodical mobile crowdsensing system from a frugality payment perspective. We explore the issue under both semi‐online (the intraperiod interactive process is synchronous while the interperiod interactive process is sequential and asynchronous during each period) and online user arrival models (the previous 2 interactive processes are sequential and asynchronous). In particular, we first propose a semi‐online frugal incentive mechanism by introducing a Lyapunov method. Moreover, we also extend it to an online frugal incentive mechanism, which satisfies the long‐term participation constraint and approximate optimality. Finally, extensive simulations show that our mechanisms satisfy the above theoretical properties. [ABSTRACT FROM AUTHOR]

Machine learning techniques require an enormous amount of high-quality data labeling for more naturally simulating human comprehension. Recently, mobile crowdsensing, as a new paradigm, makes it possible that a large number of instances can be often quickly labeled at low cost. Existing works only focus on the single labeling for supervised learning problems of traditional machine learning, where one instance associates with only label. However, in many real world applications, an instance may have more than one label. To the end, in this paper, we explore an incremental multi-labeling issue by incentivizing crowd users to label instances under the budget constraint, where each instance is composed of multiple labels. Considering both uncertainty and diversity of the number of each instance's labels, this paper proposes two mechanisms for incremental multi-labeling crowdsensing by introducing both uncertainty and diversity. Through extensive simulations, we validate their theoretical properties and evaluate the performance. [ABSTRACT FROM AUTHOR]

Ensemble selection, which aims to select a proper subset of the original whole ensemble, can be seen as a combinatorial optimization problem, and usually can achieve a pruned ensemble with better performance than the original one. Ensemble selection by greedy methods has drawn a lot of attention, and many greedy ensemble selection algorithms have been proposed, many of which focus on the design of a new evaluation measure or on the study about different search directions. It is well accepted that diversity plays a crucial role in ensemble selection methods. Many evaluation measures based on diversity have been proposed and have achieved a good success. However, most of the existing researches have neglected the substantial local optimal problem of greedy methods, which is just the central issue addressed in this paper, where a new Ensemble Selection (GraspEnS) algorithm based on Greedy Randomized Adaptive Search Procedure (GRASP) is proposed. The typical greedy ensemble selection approach is improved by the random factor incorporated into GraspEnS. Moreover, the GraspEnS algorithm realizes multi-start searching and appropriately expands the search range of the typical greedy approaches. Experimental results demonstrate that the newly devised GraspEnS algorithm is able to achieve a final pruned subensemble with comparable or better performance compared with its competitors. [ABSTRACT FROM AUTHOR]

For the demand of the collimated system improvement of a kind of helicopter, a detection algorithm for the cross line of the collimated system is proposed. Firstly, after image denoising, Kirsch edge detection is used to obtain the edge points of the cross. Then the chain code of these edge points is obtained. After analyzing the chain code, the line of the cross is detected based on a line similarity algorithm. Finally, the cross of collimated system is detected based on the combination of line feature, site feature and angle feature, which improves the reliability of the result greatly. Experiments show that the performance of the proposed algorithm is excellent, and it can meet the requirement of the practical use. [ABSTRACT FROM AUTHOR]

Deep neural networks have been successfully applied in domain adaptation which uses the labeled data of source domain to supplement useful information for target domain. Deep Adaptation Network (DAN) is one of these efficient frameworks, it utilizes Multi-Kernel Maximum Mean Discrepancy (MK-MMD) to align the feature distribution in a reproducing kernel Hilbert space. However, DAN does not perform very well in feature level transfer, and the assumption that source and target domain share classifiers is too strict in different adaptation scenarios. In this paper, we further improve the adaptability of DAN by incorporating Domain Confusion (DC) and Classifier Adaptation (CA). To achieve this, we propose a novel domain adaptation method named C2DAN. Our approach first enables Domain Confusion (DC) by using a domain discriminator for adversarial training. For Classifier Adaptation (CA), a residual block is added to the source domain classifier in order to learn the difference between source classifier and target classifier. Beyond validating our framework on the standard domain adaptation dataset office-31, we also introduce and evaluate on the Comprehensive Cars (CompCars) dataset, and the experiment results demonstrate the effectiveness of the proposed framework C2DAN. [ABSTRACT FROM AUTHOR]

The most current news recommendations are suitable for news which comes from a single news website, not for news from different heterogeneous news websites. Previous researches about news recommender systems based on different strategies have been proposed to provide news personalization services for online news readers. However, little research work has been reported on utilizing hundreds of heterogeneous news websites to provide top hot news services for group customers (e.g., government staffs). In this paper, we propose a hot news recommendation model based on Bayesian model, which is from hundreds of different news websites. In the model, we determine whether the news is hot news by calculating the joint probability of the news. We evaluate and compare our proposed recommendation model with the results of human experts on the real data sets. Experimental results demonstrate the reliability and effectiveness of our method. We also implement this model in hot news recommendation system of Hangzhou city government in year 2013, which achieves very good results. [ABSTRACT FROM AUTHOR]