Loading...

Table of Content

    28 April 2018, Volume 0 Issue 04
    #br#    Chinese Semantic Role Labeling Based on Gated Mechanism and BiLSTMCRF
    ZHANG Miaomiao, ZHANG Yujie, LIU Mingtong, XU Jinan, CHEN Yufeng
    2018, 0(04):  1.  doi:10.3969/j.issn.10062475.2018.04.001
    Asbtract ( 183 )  
    References | Related Articles | Metrics
    With the increasing research and application of deep learning, researchers have proposed a method by applying bidirectional long shortterm memory (BiLSTM) to semantic role labeling. Among them, many methods rely on a local window in word representation. In this case, the word representation may depend on the joint effect of some fixed word embeddings and it may preserve useless information in context. Aiming at this problem, we make improvements based on BiLSTM and introduce a Gated mechanism (GM) to filter information that will be fed to the next layer. In order to gain much more semantic information, we extend the depth of the BiLSTM. Due to the strong dependencies across output labels, we model tagging decisions using a conditional random fields (CRF) and transition matrices. Experiments on CPB benchmark dataset show that the F1 score of the Chinese semantic role labeling system finally improves 1.71%.
    Attribute Reduction Based on Granularity Decision Entropy
    LI Hua, JIANG Feng, YU Xu, DU Junwei, LIU Guozhu
    2018, 0(04):  7.  doi:10.3969/j.issn.10062475.2018.04.002
    Asbtract ( 196 )  
    References | Related Articles | Metrics
    In recent years, more and more attention has been paid to the attribute reduction algorithm of rough set, especially the heuristic reduction algorithm. In order to measure the attribute importance, people used different kinds of information entropy model in rough set, and obtained many reduction algorithms on the basis of the theory of information entropy to solve the problem of attribute reduction of rough set. However, there are a number of problems in the existing information entropy methods. To solve these problems, this paper firstly combines the knowledge granularity and relative decision entropy, and introduces a new information entropy model—the granularity decision entropy. Then, using the granularity decision entropy to measure the importance of attributes, the new reduction algorithm—ARGDE reduction algorithm is obtained. Finally, different UCI data sets are used to perform the experiment, and the algorithm can get better results by comparing with the existing reduction algorithms.
    A Parallel Algorithm for Mining onshelf Utility Itemset with Negative Item Values
    CHEN Lijuan, XIE Huosheng
    2018, 0(04):  13.  doi:10.3969/j.issn.10062475.2018.04.003
    Asbtract ( 166 )  
    References | Related Articles | Metrics
    In order to improve the mining efficiency of the onshelf utility itemset mining algorithms with negative item values, the paper proposed a parallel algorithm for mining onshelf utility itemset with negative item values named DTPHoun (distributed TPHoun algorithm). Based on MapReduce,  the algorithm divides the database according to the onshelf time periods. The algorithm transforms the mining work into MapReduce job, the Map phase to mine candidates in database fragments, and the Reduce phase to calculate the onshelf utility values of the candidates in parallel. The experimental results show that the DTPHoun algorithm has a good performance.
     Text Clustering Based on Improved kmeans Algorithm
    JIANG Li, XUE Shanliang
    2018, 0(04):  17.  doi:10.3969/j.issn.10062475.2018.04.004
    Asbtract ( 150 )  
    References | Related Articles | Metrics
    To solve the problem that the original kmeans algorithm is sensitive to the clustering number k, an improved kmeans algorithm is proposed. The algorithm is designed to firstly calculate the similarity between word vectors based on the principle of cooccurrence words and divides the data into k+x clusters according to the similarity threshold and then uses kmeans algorithm for k+x clusters. The proposed algorithm is applied to the text clustering. The experimental results show that the proposed algorithm is more accurate than the original algorithm.
    TSVM Learning Algorithm Based on Improved Knearest Neighbor
    LI Yu, FENG Ao, ZOU Shurong
    2018, 0(04):  22.  doi:10.3969/j.issn.10062475.2018.04.005
    Asbtract ( 115 )  
    References | Related Articles | Metrics
    (College of Computer Sciences, Chengdu University of Information Technology, Chengdu 610225, China)
    GUI Software Test Case Generation Method Based on WEHG Model
    XIANG Rifeng, MAO Yuguang
    2018, 0(04):  26.  doi:10.3969/j.issn.10062475.2018.04.006
    Asbtract ( 159 )  
    References | Related Articles | Metrics
    The graphic user interface represents the frontend in the underlying code. In software development at this stage, the GUI accounts for over 60 percent of the total code. In view of the existing model based method to generate test cases can not find software defects as soon as possible, the paper embarks from the code layer and interface layer to analyze the test program, and puts forward a WEHG GUI test model. The model features are: 1) According to the number of the defined variables and reference variables of eventhandler and the corresponding node, we can set weight’s value, thus ensuring more variable nodes generate test cases preferentially; 2) According to the definition of the event handler function, the dependency value of the dependency on the node is set up, so that the node with high dependency can be first added to the test sequence. The experimental results show that this method can detect defects in software faster, improve the detection efficiency of test cases, and reduce the cost of software testing.
    A Parallelization Method with JNI and C+〖KG-*3〗+ for Many Integrated Core Application
    SANG Zhe1, DENG Chuan1, GOU Cong1, LIU Kaixing2, BAI Mingze1
    2018, 0(04):  32.  doi:10.3969/j.issn.10062475.2018.04.007
    Asbtract ( 278 )  
    References | Related Articles | Metrics
    Currently, the Intel many integrated core (MIC) coprocessor can be used for parallel computing only through C/C+〖KG-*3〗+  and Fortran programming language, lacking the support for existed Java program. In this passage, we propose to exploit the powerful computing resources of MIC coprocessor to boost the Java application basing on the hybrid parallel computing strategy of Java native interface (JNI) technique and C+〖KG-*3〗+, the interface achieves the data exchange between C+〖KG-*3〗+ and Java program. We design an experiment to test and analyze MICbased Java multithreads parallel computing program. The results show the performance improvement of Java program brought by Phi coprocessor Java program.
    Moving Object Detection Based on NMF and Similarity Analysis
    FAN Xinnan, XUE Ruiyang, SHI Pengfei, LI Min, NI Jianjun
    2018, 0(04):  37.  doi:10.3969/j.issn.10062475.2018.04.008
    Asbtract ( 137 )  
    References | Related Articles | Metrics
     An algorithm of moving object detection fusing nonnegative matrix factorization (NMF) and vector similarity analysis is proposed. Firstly, the background is reconstructed from the continuous image sequence by using the modified NMF algorithm. Then, the similarity between the detected pixel and the recovered background model is analyzed, and the background and foreground are distinguished according to the similarity. In order to reduce the amount of computation and reduce the interference of dynamic background to the detection results, the method of kernel density estimation (KDE) is used to estimate the motion area before the similarity analysis is performed. The experimental results show that the proposed algorithm can recover the background image more accurately and detect the moving object effectively.
    Underwater Image Restoration Based on Color Attenuation Prior and White Balance
    HAN Hui1, ZHOU Yan1,2, CAI Chendong1
    2018, 0(04):  42.  doi:10.3969/j.issn.10062475.2018.04.009
    Asbtract ( 230 )  
    References | Related Articles | Metrics
    Due to the absorption and scattering of light by floating particles in water, and the different attenuation degrees of light at different wavelengths under water, the underwater images generally have the problems of fuzzy details, low contrast and color distortion. Aiming to improve the quality of underwater images, an underwater image restoration method based on color attenuation prior and white balance is proposed. Firstly, the scene depth map is obtained according to the color attenuation prior in HSV color space of the image. Secondly, the underwater optical attenuation characteristic is used to estimate the corresponding background light intensity and underwater transmittance of RGB channels in order to realize image clarity. Finally, the improved white balance method is adopted for the color correction of the clear underwater image. The experimental results show that the proposed method can significantly improve the detail sharpness and color fidelity of underwater images, and the visual effects are more similar to images captured in natural scenes.
    SCS: A Model of Saliency Detection Based on Deep Learning
    ZHANG Hongtao, LU Hongying, LIU Tengfei, ZHANG Lingyu, ZHANG Xiaoming
    2018, 0(04):  48.  doi:10.3969/j.issn.10062475.2018.04.010
    Asbtract ( 186 )  
    References | Related Articles | Metrics
    In this paper, we propose a method of saliency detection based on deep learning. This method extracts the lowlevel contrast features and highlevel semantic features involved in the two visual attention mechanisms, and combines both of them to obtain a classificationbased saliency detection model SCS. Through the comparison experiment, it is concluded that the proposed detection model has significant advantages in the accuracy of saliency detection.
    Indoor Window Detection Based on Image Contour Analysis
    KONG Qianqian, ZHAO Liaoying, ZHANG Li
    2018, 0(04):  56.  doi:10.3969/j.issn.10062475.2018.04.011
    Asbtract ( 179 )  
    References | Related Articles | Metrics
    Aiming to the problem of indoor window detection, this paper proposed an approach of indoor window detection based on image contour analysis. The image is preprocessed and a binary image is obtained by threshold segmentation and morphological processing. Then, the contours of the image are extracted by the border tracking based on the topology structure analysis, and stored as the sequence points. The contours of the qualifying conditions are selected according to the characteristics of the window contours. The minimum enclosing rectangles of each contour are figured up and the shortest distance between each two rectangles are calculated. Finally, the rectangles are classified and merged by the minimum spanning tree to obtain the position of the window. Experiment results show that the proposed method can effectively detect the windows in different indoor scenes.
    Gesture Recognition Method Based on Deformable Convolution Neural Network
    SU Junxiong, JIAN Xueting, LIU Wei, HUA Junda, ZHANG Shengxiang
    2018, 0(04):  62.  doi:10.3969/j.issn.10062475.2018.04.012
    Asbtract ( 179 )  
    References | Related Articles | Metrics
    Convolution neural network itself has a rich ability of expressing features and learning, but in essence, the module geometric transformation ability is fixed. Therefore, the VGG16 network structure is improved by introducing a deformable convolution kernel, and a convolution neural network structure named DCVGG is built to study the gesture recognition. In different data sets, the gesture recognition method based on deformable convolution neural network can input RGB image data directly into the network. The results show that the average recognition rate of gestures is over 97%, which can improve the performance of the network, enhance the tolerance and diversity of the convolution neural network to the sample object, and enrich the expression ability of the convolution neural network. Compared with the traditional LeNet5, VGG16 structure and traditional feature extraction by hand, DCVGG is deeper than the traditional structure, the robustness is better, the recognition rate is stronger, which can provide reference for the effective recognition of gestures in complex background, and has some extension ability.
    Power Failure Sensitivity Prediction Algorithm Using Ksupport Sparse Logistic Regression
    GENG Juncheng1, ZHANG Xiaofei1, SUN Yubao2, WU Bo1, ZHOU Qiang2
    2018, 0(04):  68.  doi:10.3969/j.issn.10062475.2018.04.013
    Asbtract ( 144 )  
    References | Related Articles | Metrics
    The prediction of customers with high sensitivity of electric power failure can provide data and decision support for the electric power service departments to offer precision marketing and differentiated services. With regard to the electric power failure sensitivity problem, we propose the electric power failure sensitivity assessment algorithm using ksupport norm regularized logistic regression. Different from the normal l1 norm, ksupport norm is the tighter convex relaxation of l0 norm on the Euclidean norm unit ball and able to select multiple correlated variables to predict the response, which can promote the accuracy of predicted results. Firstly, the variables or factors for predicting response are selected from multiple aspects including the customer information, electric consuming information, electrical bill information, 95598 work sheet, power failure events, etc. The sample set is constructed by collecting the variable information of each consumer. Secondly, ksupport norm regularized logistic regression model is used to predict customers with high sensitivity of electric failure. In terms of forwardbackward operator splitting, an iterative optimization algorithm is also proposed to decompose the original problem into two subproblems and solve the model effectively. Furthermore, dominance analysis method is adopted to identify the importance of each variable for predicting the response result. The model is validated by using about one million customer data from a province supply board and has good prediction accuracy. The experimental results demonstrate the effectiveness of our prediction model.
    GeoDAE for Pointofinterest Recommendation
    ZHANG Wenxiang
    2018, 0(04):  74.  doi:10.3969/j.issn.10062475.2018.04.014
    Asbtract ( 102 )  
    References | Related Articles | Metrics
     Personalized pointofinterest (POI) recommendation is crucial to the development of locationbased social networks (LBSNs). It not only helps users explore new places but also enables thirdparty services to better provide service. Previous studies on this topic treat all POIs as equal. Learning preferences within category makes sense, but the scale in which the frequency of checkins operates is not comparable across categories. In this paper, we transform the checkin frequency into categorybased preference according to TFIDF theory. And then we propose a GeoDAE model for the geographical proximity among POIs. The experimental results based on datasets from realworld LBSNs show that the proposed model achieves better performance than other stateoftheart methods, and the proposed model is a better alternative for POI recommendation.
    Detection of Potential Audit Doubts Based on Improved Leaders Operator
    SHAO Jinwei1, LIN Jun2, LIU Yating3, XIAO Jiali4
    2018, 0(04):  79.  doi:10.3969/j.issn.10062475.2018.04.015
    Asbtract ( 99 )  
    References | Related Articles | Metrics
     Database query method audit doubts discovery relies on the prior knowledge of the auditors, but when the auditors have not enough audit experience and the amount of audit data is huge, it is difficult to take advantage of big data and find the audit doubts from the massive data. And so, in order to solve this problem, a method based on improved Leaders operator and iterative clustering is proposed. In the absence of prior knowledge, Leaders algorithm is used for automatic initial clustering of large audit data, and then, the random sampling fusion method is introduced to optimize the clustering results based on that initial clustering center, finally, the multiple iterative clustering method is used to further find the small clusters with fewer or doubtful instances and thus the accurate clustering of large audit data is achieved. The data clusters with fewer instances or obviously abnormal behavior are identified as potential audit doubts, which can cooperate with audit experience to assist auditors to locate audit doubtful points quickly and accurately. Experimental results verify the effectiveness of the proposed algorithm, and show that the proposed algorithm is helpful to find out the audit doubts from the mass data, narrow the scope of doubts screening and improve the audit efficiency.
    Text Similarity Calculation Method Based on Levenshtein and TFRSF
    ZANG Runqiang1, SUN Hongguang1,2, YANG Fengqin1,2, FENG Guozhong1,2, YIN Liangliang1
    2018, 0(04):  84.  doi:10.3969/j.issn.10062475.2018.04.016
    Asbtract ( 141 )  
    References | Related Articles | Metrics
     Finding and collecting personal information in social networks can establish the information system with the curriculum vitae, life, hobbies, friends and the other attributes. But there are lots of same name users in different social networks. In order to solve the ambiguity of the same name, we calculate the user information similarity to decide whether they belong to the same person. If the information describing the document position is reversed, it will lead to computer misjudgment. In order to solve this problem, the Levenshtein and TFRSF methods are used to calculate the word frequency and edit distance to judge whether the attribute values are the same. The experimental results show that the proposed method of calculating the similarity of texts improves the accuracy of various evaluation indexes. The precision, recall and F1 of this method are more than 87%.
    Vocabulary Semantic Similarity Computation Based on HowNet and Search Engine
    WU Kejie, WANG Jiawei
    2018, 0(04):  90.  doi:10.3969/j.issn.10062475.2018.04.017
    Asbtract ( 237 )  
    References | Related Articles | Metrics
    This paper proposes a method of computing lexical semantic similarity based on HowNet and search engines. The similarity computation is optimized by using the depth, density and information of semantic primitive in the hierarchy tree. The search engine based lexical semantic similarity computation is optimized by combining the point by point common information (PMI) algorithm with the normalized Google distance (NGD) algorithm. The lexical part of speech is used as weighting factor to merge the word similarity computation between HowNet and search engine. The experimental results show that, compared with the semantic similarity calculation method based on HowNet and search engine, the average similarity of the proposed method on NLPCC test set is closer to the evaluation criteria of the test set, and lexical similarity in the car ticket calculation fields has a good application effect.
    Information Security Risk Assessment Based on Improved Bayesian Network Model
    HUANG Yujie, TANG Zuoqi
    2018, 0(04):  95.  doi:10.3969/j.issn.10062475.2018.04.018
    Asbtract ( 265 )  
    References | Related Articles | Metrics
    With the advent of the information age, information security issues become increasingly complex and diverse, so a costeffective solution should be badly in need. Based on the previous research, this paper further improves the application of Bayesian network model in information security risk assessment. Firstly, it analyzes the types of risk elements of information system, and puts forward a new method to determine the risk factors, that is, the common relationship between factors. Then, the information system index system is determined according to the factor relation. Combined with the conditional probability of experience accumulation, the Matlab Bayesian network toolbox (BNT) is used to construct a complete Bayesian network risk assessment model, which includes the analysis of the assessment process, the use of methods and the determination of risk levels. Finally, by analyzing the improved Bayesian assessment model, the probability of each level of risk is deduced according to experimental data. The simulation results are consistent with the actual results, which show that the improved evaluation method is effective and reasonable.
    Swarm Robotic System Selfhealing Method Based on #br# Improved Immuneinspired Swarm Aggregation Algorithm
    LIU Yangju, NI Jianjun
    2018, 0(04):  100.  doi:10.3969/j.issn.10062475.2018.04.019
    Asbtract ( 115 )  
    References | Related Articles | Metrics
    Swarm robotic system is an important research direction of multirobot system, which is composed of many simple and undifferentiated robots. Compared with the individual robot, swarm robotic system has good fault tolerance and robustness, but it will be affected when the partial failure of a robot appears, namely a robot has information interaction ability but no drive ability. Aiming at this problem, an optimal recovery strategy is proposed, which is based on the granuloma formation algorithm inspired by immune system. In the proposed approach, a discrete particle swarm optimization algorithm is introduced, which makes the swarm robotic system have fault selfhealing capability and complete its task quickly and effectively. The experimental results show that the proposed approach has good effects in the swarm robotic selfhealing system.
    Robot Target Searching Method Based on Improved Biologically Inspired Neural Network
    ZHANG Zhitong, NI Jianjun, MO Zhengpei
    2018, 0(04):  106.  doi:10.3969/j.issn.10062475.2018.04.020
    Asbtract ( 131 )  
    References | Related Articles | Metrics
    Aiming at the problem of robot target search in an unknown environment, the search area is divided according to the robot’s ability. The target points will leave pheromones in the local range during their motion, and these pheromones will decrease with time. The robot can detect the size of these pheromones and affect the next robot’s location. In this paper, the active value in the exploration range of robot is selected by improving the biologically inspired neural network. In order to prevent multiple choices of the same point in a continuous time period, a tabu search is introduced, and the same points are selected in the tabu table, which can be effectively prevented from getting into the local best advantage. Compared with the random search method, the method is proved to have a good effect on the target search.
    Construction and Optimization Implementation of 3D Radar Detection Range Based on OSG
    REN Fei, WANG Jiarun, LI Xiaojuan, YIN Hui
    2018, 0(04):  111.  doi:10.3969/j.issn.10062475.2018.04.021
    Asbtract ( 248 )  
    References | Related Articles | Metrics
    To help the commander determine situation accurately in electronic countermeasure, this paper researches the construction and optimization implementation technology of 3D radar detection range. Based on Open Scene Graph (OSG) 3D graphics engine, and combining radar detection 2D area with vertical detecting distance calculating method, this paper achieves the general method for calculating 3D radar detection range. In the light of the large volume of radar data and realtime data updating, it is optimized in two stages: reducing the algorithm complexity and optimizing calculation method; using OSG multithreading technology to conduct parallel processing for the calculating drawing process. Through the test and analysis, the efficiency of multithread method is proved.
    Progression Prediction Model of Chronic Kidney Disease Based on  Decision Tree Ant Path Optimization and Logistic Regression
    FENG Miao1, QI Xiaorong2, LI Zhi1
    2018, 0(04):  117.  doi: 10.3969/j.issn.10062475.2018.04.022
    Asbtract ( 116 )  
    References | Related Articles | Metrics
    Chronic kidney disease (CKD) is a progressive disease, it will lead to the development of the disease and even renal failure if not treated in a timely manner. To study the progression probability from earlystage to endstage of CKD patients, a prediction model of CKD progression probability is proposed. Combining decision tree ant path optimization (DTAPO) and logistic regression (LR) algorithm, this paper divides CKD patients’ data into two categories: progress (P) and non progress (NP), the classification accuracy rate and recall rate are obtained so as to calculate the probability from the stage 3 to the stage 4 or 5. It is demonstrated from the experimental results that when the number of features is 13, the prediction algorithm combining decision tree ant path optimization algorithm with logistic regression achieves the best performance, and the accuracy rate of classification is 98.84%. The probability of progression from the stage 3 to the stage 4 or 5 is 0.9827.
    Price Prediction of Secondhand House Based on ARIMA Model
    ZHENG Yongkun, LIU Chun
    2018, 0(04):  122.  doi:10.3969/j.issn.10062475.2018.04.023
    Asbtract ( 183 )  
    References | Related Articles | Metrics
    In recent years, house prices in China’s firsttier and secondtier cities have continued to rise, and houses have become a hot topic of discussion in daily life, we have to guess the future trend of housing prices. In this paper, we take up the average price of secondhand housing in Guangzhou and Shenzhen since 2013 from a domestic wellknown largescale real estate websites, use the ARIMA model to rollingly forecast the future housing price, and use RMSE to judge the prediction accuracy. The results show that the model can predict the average price of secondhand house continuously, and the prediction accuracy is higher, which can be used as reference for housing traders.