Search

Existing NL2SQL approaches do not fully utilize the information of data table columns, which plays an important role in the semantic understanding of the problem and the logical generation of SQL statements. This paper proposed an NL2SQL method that fused data table columns (FC-SQL) to improve the overall accuracy of SQL generation. Firstly, this method utilized BERT to merge the problem and database table columns for encoded representations. Secondly, it used multitask learning approach to construct a multi-task network by combining parallel and cascade to predict different sub-tasks. Finally, for the conditional value extraction sub-task, this method computed the similarity between the words in the problem and the table columns by fusing the information of the columns, and it used the similarity value as a weight to compute each word as the conditional value probability of each word as a conditional value, thus improving the accuracy of conditional value prediction. The logical form accuracy and SQL execution accuracy on the TableQA dataset reach 88.23% and 91.65%, respectively. This paper designed ablation experiments to verify the effect of table columns information on the model. The experimental results show that the incorporation of table columns improves the effectiveness of the conditional value extraction sub-task, which in turn improves the overall accuracy of the NL2SQL task and provides better SQL generation compared to the comparison model. [ABSTRACT FROM AUTHOR]

Indexing is an important method to improve database performance. At present, with the development of reinforcement learning algorithm, there are a series of methods to solve the index recommendation problem（ISP） by reinforcement learning. Aiming at the problem that the existing deep reinforcement learning index recommendation algorithm has long training time and unstable training, this paper proposes an index recommendation algorithm based on A2C（Advantage Actor-Critical）, PRELIA（Parallel Compensation Learning Index Advisor）. In order to improve the accuracy and efficiency of index selection and reduce the occupation of index space, the algorithm adds the characteristic matrix of the number of rows scanned by load index and normalizes the reward value. Experimental results on different data sets show that the proposed algorithm can guarantee the index recommendation quality equivalent to that of the compared algorithm, while the recommended index occupies less storage space, and the training time is more than 4 times longer than that of the baseline algorithm. [ABSTRACT FROM AUTHOR]

Aiming at the problem that LF ant colony clustering algorithm does not distinguish the importance of data set attributes, the algorithm efficiency is low and the clustering effect is unstable, this paper proposed a weighted global memory ant colony optimization (WGACO) based on entropy weight. Firstly, it calculated the entropy weight of each attribute by entropy weight method, and modified the Euclidean distance calculation formula to improve the clustering accuracy. It initialized the data object with the value of the largest weight attribute to enhance the stability of the clustering effect, and introduced the global memory matrix to reduce the invalid movement of ants and improve the efficiency of the algorithm. It added the convergence conditions of the algorithm to improve the practicability of the algorithm. Selecting seven real data sets and three artificially generated data sets in UCI database for numerical experiments, and compared WGACO with GMACO,SMACC and ILFACC three improved LF algorithms. The experimental results show that WGACO has a relatively good improvement in accuracy, algorithm efficiency and stability, and also has a good performance in processing high-dimensional data. Finally, WGACO has performed well in the segmentation of mall member users, reflecting its practical value. [ABSTRACT FROM AUTHOR]

The multifractal theory only performs statistical analysis on the singular exponents observed at any point on the geometric support of the fractal body. Multifractal correlation studies the spatial correlation characteristics between two points with different singular exponents. It is a generalization of single point statistics in multifractals. The two characteristics have a certain degree of complementarity. This paper investigated a method that combined the multifractal and its correlation features in the electrocardiogram（ECG） signal classification. Firstly, the method analyzed the multifractal and correlation characteristics of four types of ECG signals and obtained their respective features. Then it combined the probability distribution to select appropriate features for combination. The combined features fed into a support vector machine（SVM） for classification. This method tested on the MIT-BIH arrhythmia database and achieved an average accuracy of 97.90% after 20 training tests. Compared to independently using multifractal features, this method significantly improves the classification accuracy. [ABSTRACT FROM AUTHOR]

Spatial direction relation is a basic concept in spatial cognition, which plays an important role in the fields of spatial database, artificial intelligence, robot and so on. With the application of spatial direction relation, the qualitative reasoning with spatial direction relations has gradually become a hot issue, which has received a lot of attention. In order to further introduce and explore the research status of qualitative spatial reasoning with directions. Firstly, we combed and summarized the existing models for reasoning with directions in detail, compared and analyzed all kinds of models, and the pointed out the characteristics and the applicable scope of all kinds of models. Then, we systematically surveyed the researches of spatial reasoning with direction relations including composing, inversion, consistence checking with direction relations for two-dimensional, three-dimensional and uncertain objects and the researches of compound spatial reasoning based on direction, topology and distance. Finally, the current challenges and potential development trends are also discussed. [ABSTRACT FROM AUTHOR]

Traditional relational databases use the artificial way to achieve index recommendations that can no longer meet the growing requirement of data, and using machine learning technology can effectively solve the index optimization problem of databases. The limitation that only recommended the index in a static database and couldn't update the index configuration in time, this paper proposed a multi-attribute index intelligent recommendation approach (MIRA) to recommend multi attribute index configurations for a set of workloads using reinforcement learning under databases with dynamically changing data. Experimental results on TPC-H datasets show that MIRA can effectively recommend the optimal index configuration for workload, and outperforms the defined comparison baselines and related reinforcement learning methods. [ABSTRACT FROM AUTHOR]

Wireless communication and mobile application put forward higher requirement for mobile RFID authentication. Aiming at the security and efficiency problem of the existing authentication protocol, this paper proposed a new mobile RFID high-efficiency authentication protocol. The protocol protected the security of information transmission by using arithmetic operation, bit-wise operations, permutation and combination of bits data. It defended the illegal attack by adversary through the progressive verification of random number by database and mobile reader, as well as bundled verification by the tag. For one thing, as the communication entity performed light-weight operations, the amount of calculation was greatly reduced. For another, as the communication entity initially verified the authenticity and legitimacy of random data in time, it reduced time of the invalid authentication. These two points showed that the protocol can improve the authentication efficiency. This paper proved the correctness of the protocol using GNY logic formal language. Analysis results show that the protocol can resist different types of attacks. Simulation results show that the protocol has high authentication efficiency as compared with similar mobile RFID authentication protocols. The protocol has better application value. [ABSTRACT FROM AUTHOR]

Aiming at the problem that isolation forest is unreliable in the face of high-dimensional data by randomly selecting attributes, this paper proposed an improved anomaly detection algorithm high contrast subspace isolation forest (HiForest). Firstly, this algorithm selected the subspace with high contrast value based on the deviation between the edge probability and the joint probability of each attribute in the subspace. After that, the algorithm constructed isolation trees which had a stronger ability to isolate outliers in the relevant subspace. Finally, the algorithm integrated these isolation trees into isolation forest, and obtained the anomaly score by traversing the average path length of data points in the isolation forest. Compared with traditional anomaly detection algorithms, the experimental results based on the ODDS database show that the HiForest algorithm improves the AUC, accuracy, recall rate significantly. Therefore, HiForest algorithm is an anomaly detection algorithm suitable for high-dimensional datasets with higher detection accuracy. [ABSTRACT FROM AUTHOR]

For precise classification of pathological breast WSI images, the study proposed the gated convolution network based on hybrid connection（HC-GCN）, set up a hybrid block of local residual connection and global dense connection, and through embedding the Squeeze-Excitation-and-Gated（SEG） module into the hybrid block, established a backbone network for alternate connection between the hybrid block and the transition layer. In combination with a training model for image data enhancement based on quad-tree image segmentation method, the experimental results based on the BreastSet clinical data set showed 92.24%, 92.83% and 92.18% of accuracy at the image level, patient level and pathology level respectively. Therefore, compared with other methods, this method is of great clinical application value for improved accuracy as well as reduced number of parameters and amount of computation. [ABSTRACT FROM AUTHOR]

Most of the existing multi-label learning techniques consider the correlation learning problem but ignore the inconsistency feature of structural properties, which can change the structural property of the label space obtained by function mapping and it can directly affect the classification performance. In order to deal with this problem, this paper proposed a multi-label classification algorithm based on structural property preserving policy and correlation learning policies. Firstly, the algorithm constructed a mapping function to realize the mapping between the feature space and the label space. Secondly, it used structural property preserving strategy based on feature data to reduce the difference between feature data caused by linear transformation. Finally, this paper introduced the correlation learning strategy based on label data to further optimize the parameters of the proposed algorithm to improve the classification performance. The experimental results through a group of benchmark instances show that compared with the other classical multi-label classification algorithm, the proposed algorithm is much superior in classification performance, and the algorithm has great effectiveness. [ABSTRACT FROM AUTHOR]

A large number of low-frequency words existed in Chinese short text classification, reasonable development and utilization could effectively improve the effect of text classification. Aiming at the problem that low-frequency words couldn't be effectively used in the research of text classification based on word vector, this paper proposed a method of data enhancement based on low-frequency words. Firstly, it used the data gene rated by the restricted text generation model to fine tune the word vector of low-frequency words, and then used a word vector construction algorithm to transfer the updated information of high-frequency words to low-frequency words, so that low-frequency words could obtain a more accurate word vector representation that conformed to the distribution of training set. Secondly, it introduced prior knowledge such as similar words and entity concepts to supplement context information. Finally, it used the improved Chi-square statistics to remove the obvious noise words, and designed the word attention layer to weight each word to reduce the impact of irrelevant noise on classification. Experiments on several basic classification models show that each basic model has been significantly improved after applying the improved method. It not only shows the effectiveness of the proposed method, but also shows that low-frequency words can improve the classification effect in short text classification task. [ABSTRACT FROM AUTHOR]

When calculating similarity, the existing differential privacy recommendation algorithms directly calculate based on user-scheme data, but ignore the influence of scheme attributes on user preferences, fail to reflect the real preferences of users, and cannot make accurate recommendations. To solve this problem, this paper proposed differential privacy recommendation method considering user interest analysis. In this method, it collected firstly users' interest ratings for scheme attributes, and secondly clustered the user-scheme attributes rating data by K-means + +. Then, it used the differential privacy algorithm to select the nearest users, and recommended suitable schemes for the target user. Finally, taking the recommendation of nursing home schemes as an example, the experimental results show that compared with KDPC, DPCF and PNCF, the proposed algorithm can reduce the mean absolute error by about 19. 0%, 34. 0% and 37. 7% under the same privacy budget. The mean absolute error decreases by 10. 4%, 20. 3 % and 21. 4 % for the same size of the nearest neighbor set. Therefore, based on protecting the privacy of users, the algorithm further improves the recommendation accuracy. [ABSTRACT FROM AUTHOR]

Representation-based classification ( RC) usually uses all the training samples of all pattern classes to represent a test sample . However, whether it is necessary to use all pattern classes to represent the test sample remains to be investigated . So, this paper proposed two-stage representation-based classification framework. First, it used a RC algorithm to calculate the representation coefficients of the test sample w. r. t the training samples of all classes, and found the first k ( k≥1) classes with the minimum representation error. Then with the training samples of these k classes, it applied a RC algorithm again to represent the test sample, based on which, the label of the test sample was finally determined by finding the minimal representation error class among those k classes. In addition, this paper developed a nonnegative weighted collaborative RC ( NWCRC) algorithm. The first and second RC algorithm of the proposed RC framework could be the same or different. Setting the first and second RC algorithms to be the same, with five different RC, experiments were carried out on five databases, experimental results of which show that the proposed two -stage RC framework can significantly improve the classification accuracy of the original RC algorithms in most cases. [ABSTRACT FROM AUTHOR]

Aiming at the ubiquitous network spammers on e-commerce platform, this paper proposed an algorithm considering network structure and time characteristics to detect the spammer groups in the comment network. The algorithm consisted of four steps: a) mining the target products that were vulnerahle to attack by the spammers based on the analysis of the structural characteristics of the comment network; b) this paper proposed an algorithm for mining the suspicious period when the spammer groups attacked the target product inspired by the "co-bursting phenomenon" ; c) this paper constructed the induced subgraph of target products-reviewers based on the data of target product in suspicious period, and applied hierarchical agglomerative clustering algorithm to generate candidate spammer groups on the subgraph; d) in order to filter out the normal users who shopped and commented during the suspicious period, this paper proposed a spammer groups purification method, and then classified the purified groups based on the behavior characteristics of the reviewers. The experimental results based on real data sets show that the proposed algorithm can accurately and efficiently detect the network spammer groups active on e-commerce websites. [ABSTRACT FROM AUTHOR]

Aiming at the problems that principal component analysis was unable to use the class information of the samples and the direction of the maximum principal component was likely to cause overlap of the samples from different classes, this paper proposed a feature transformation method fusing minimum principal component subspaces of different classes. Firstly, the proposed method analyzed the component of each class of data ; and then it fused the projection results of the original data in all subspaces to form a new feature space ; finally, this paper used the KEEL benchmark database to test the method. The results show that the proposed method can construct a feature space that is more conducive to classification compared with other feature transformation methods, which can help to improve the classification accuracy of the classifier. [ABSTRACT FROM AUTHOR]

Aiming at the problems of sensitivity of initial center, high communication overhead of nodes and low efficiency of cluster in big data clustering algorithm based on partitioning, this paper proposed a partitioning-based clustering algorithm using grid density and locality sensitive hash function based on MapReduce, named PBGDLSH-MR. Firstly, based on the initial data set, it proposed the GDS(grid density strategy) to get the initial clustering center, which avoided the sensitivity of initial center caused by random selection of initial cluster center. Secondly, it proposed the DP-LSH ( data partitioning based on locality sensitive hash functions) to map more closely related data objects into the same subdataset and get data partitions on the map. Meanwhile, it designed a formula SI( similarity improvement) to evaluate the data partitioning results, reduced the communication overhead between nodes. In addition, this paper designed an AGS (adaptive grouping strategy) to handle data skew in data partitions, which improved the cluster efficiency. Finally, based on MapReduce, it mined the cluster centers in parallel to gene rate the final clustering results. The experimental results show that the PBGDLSH-MR has better clustering results and performs better parallelization in big data. [ABSTRACT FROM AUTHOR]

At present, microarray chip technology and sodium bisulfite sequencing technology have contributed a large amount of DNA methylation data. Based on different data type, there have been many methods to identify differentially methylated locis and regions. To sort out the methods of differentially methylated pattern identification, this paper Cindy introduced the research status of DNA methylation, including detection methods and data types of DNA methylation, as well as two kinds of differentially methylated patterns. Then the paper elaborated on the identification methods of differentially methylated locis and regions on the chip data,and the identification methods of differentially methylated patterns on the sequencing data based on eight different algorithm principles. This paper focused on the principles, application scenarios, advantages and limitations of these algorithms. Finally, the paper pointed out the problems of differentially methylated pattern identification and possible future development trends. [ABSTRACT FROM AUTHOR]

This paper considered a wireless communication network with multiple UAV base stations and multiple users.In the network,due to the limited spectrum resource,the backhaul links of the UAV base stations and the data links of the users shared the same spectrum.In order to use the spectrum resource efficiently and improve the communication performance of the users,this paper jointly optimized the bandwidth and power of the backhaul and data links as well as the flight trajectory of the UAV base stations,to maximize the minimum average rate of all users.This joint optimization was subject to the UAVs’ mobility,the total spectrum bandwidth,and the total transmit power,and the involved joint optimization problem was non-convex,thus it was difficult to find the optimal solution.This paper proposed an efficient algorithm to obtain a high-quality suboptimal solution of it.Simulation results show that the minimum user rate achieved by the proposed joint optimization algorithm is significantly higher than that of the benchmark schemes. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the embedded capacity and security of the single-carrier information hiding algorithm can not be further improved because of the number of carriers, combining the carrier preprocessing and the embedding of secret information with the image angle structure feature, this paper proposed a multi-carrier information hiding algorithm based on the angle structure feature( ASD) . Firstly, this paper used the angular structure features of the extracted image to classify the carrier set, and obtained the image angle structure feature direction field coding. Secondly, this paper segmented the secret information according to the carrier classification number and the preprocessing such as scrambling and optimization. Finally, based on the coded data of the ASD direction field, this paper performed information hiding on a plurality of types of carriers according to the information hiding rule. In particular, this paper designed the judgement selection option during secret information extraction to further improve the performance of the algorithm. The experimental analysis shows that the algorithm has good concealment, strong anti-analysis ability and robustness, and is suitable for applications such as large-capacity secret information covert communication with high security requirements. [ABSTRACT FROM AUTHOR]

This paper provided a comprehensive and structured survey by analyzing the most advanced methods of HUPM. First of all, this paper introduced the relevant concepts and formulas of HUPM and gave application examples of HUPM. It classified and analyzed the most common and advanced key technologies used to mine different types of HUPM, including A priori-based, tree-based, list-based, projection-based, vertical/horizontal data-based, index-based, and more. Because the static data is difficult to meet the actual needs, this paper summarized the HUPM methods applied on the data stream, mainly including based on the incremental methods, based on the sliding window model methods, based on the time decay model methods, based on the landmark model methods and so on. Finally, this paper gave the shortcomings of the current technologies and the direction of improvement, and proposed new research methods in a targeted manner. [ABSTRACT FROM AUTHOR]

In order to obtain the impact of some events on the final system fault based on the cumulative data of existing system fault events, this paper proposed a potential fault probability analysis method based on space fault network. The feature of this method was to establish event relational database and draw SFN based on the various events and the relationships between events during the system operation. When some events had happened under certain conditions, according to the causal logic and transmission probability of these events, it could get whether these events could cause system fault, fault mode and occurrence possibility. It gave a simple example to illustrate the application process and analysis effect of the method. The method can be applied to large-scale fault data processing and fault potential analysis in the system fault evolution process. [ABSTRACT FROM AUTHOR]

According to the shortcoming that the low accuracy during detect the stops of sampling irregular trajectory, this paper proposed an algorithm based on the time series clustering to find the stops in trajectories. Firstly, based on the data field theory, this work designed a hybrid feature density measurement method considering the spatial and temporal characteristics. Then, this method used the filtering-refining strategy to extract stops based on the feature that the center of stop was denser than the entrance. In the filtering phase, some points that were continuous in time and met the minimum density threshold was taken as back up stops. During the refining phase, it filtered true stops according to the maximum threshold. The experimental results show that the proposed method can effectively detect the staying points in the sampling irregular trajectory. It not only has lower time consumption than the existing methods, also has higher accuracy. [ABSTRACT FROM AUTHOR]

Aiming at the problem of large amount of data and high processing cost in WSN. This paper proposed a large data technology based on SON and multicast data stream routing technology. The method implemented the large data technology in WSN and calculated its communication cost. Firstly, it loaded the MapReduce interface dynamically into the appropriate nodes of the WSN. Secondly, it implemented map and reduced functions in the network by embedding the enhanced SON-WISE protocol in the controller and sensor nodes. Then, it evaluated the optimal deployment scheme of MapReduce by the multicast routing technology in the network. Next, it integrated the above functions and components through the operating system Contiki, implemented large data technology in WSN and calculated the corresponding cost. Finally, this paper verified the proposed method by implementing large data processing in WSN under different constraints and calculating communication cost. The results show that the proposed method is economical. [ABSTRACT FROM AUTHOR]

Aiming at the problems of low accuracy and poor security in existing fingerprint template generation algorithms, this paper proposed a fingerprint template generation algorithm based on adaptive projection. The algorithm vertically projected the distance between the pairs of effective minutiae points onto the adaptive circle and quantized the projected vector to obtain a bit string. In addition, in order to enhance the security and privacy of the data, the proposed algorithm performed XOR and random index scrambling transformation on the bit string to finally generated a fingerprint template. The experimental results in the database FVC2002 DBl and DB2 show that the proposed algorithm has higher performance of accuracy and security than the related algorithms. [ABSTRACT FROM AUTHOR]

When there exist mixed contaminations in face images, traditional low-rank recovery algorithms usually imposes only one constraint on the corresponding contaminations, it cannot recover clean samples very well. In order to solve this problem, this paper proposed a structured robust low-rank recovery algorithm ( SRLRR) . The SRLRR algorithm imposed low-rank constraint on the 2D error image to remove the continuous contamination, and introduced sparse constraint to separate the noise that obeyed the Laplacian distribution in samples. Moreover, the proposed algorithm imposed a block-diagonal structured constraint on the representation coefficient to learn the more discriminative low-rank representation. The experimental results on three commonly and using standard databases verify the effectiveness and robustness of the proposed SRLRR algorithm. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the attacker can modify original training data to generate poisoned data to poison the machine learning model in the process of training the model, this paper proposed a poisoned data detection method based on data complexity. On the basis of the normal data set, the method poisoned the sample instances in the normal data set based on a direct gradient ascent strategy, and exploited the influence of the poisoned data on the data complexity of the normal data set to build a detection model that could identify the poisoned data. The detection accuracy of this method in selected application scenarios was better than the existing method. The experimental results show that the poisoned data can effectively reduce the predictive ability of the machine learning model, and the application of the method based on data complexity can effectively detect the poisoning data and reduce the adverse effects of the poisoned data on the model prediction ability. [ABSTRACT FROM AUTHOR]

In order to increase the accuracy rate of signature verification for higher-level forgery, this paper developed a dynamic signature verification method based on timing feature fusion. First, during the writing process, it created a dynamic timeline, and extracted two timing features, the handwriting pressure and writing speed. Then it designed a fusion model of timing features based on the data of the features above. This paper calculated the similarity between the test signature and the sample signature by a similarity measure method of multidimensional spatial model. Finally, it realized signature verification by the analysis of similarity. The experiments show that the proposed algorithm has the advantages of high accuracy rate and good universality. [ABSTRACT FROM AUTHOR]

This paper proposed an improved residual network (ResNet) expression recognition algorithm. The algorithm used small convolution kernels and a deep network structure to solve the problem of accuracy reduction with the increase of depth by the residual module. The experiment overcame the shortcoming of insufficient data through transfer learning, which could effectively prevent overfitting. The network architecture used a linear SVM for classification. The experiment used the ImageNet database to pre-train network parameters to have an excellent ability to extract feature. According to transfer learning, the algorithm used the FER-2013 database and the expanded CK + database to fine-tune and train network parameters, and overcame the problem that shallow networks rely on manual features and deep networks were difficult to train. The results show the recognition rates is 91.333% and 95.775% on the CK + database and the GENKI-4K database,respectively. The classification accuracy of SVM in CK + database is about 1% higher than that of softmax. [ABSTRACT FROM AUTHOR]

Aiming at the problem that lightweight deep neural network MobileNet can reduce classification accuracy, this paper proposed a D-MobileNet( dilated convolution MobileNet) model based on local receptive field expansion by introducing dilated convolution kernel into a convolution layer of MobileNet model. The models consisted of three structures according to the location of the dilated convolution kernel. Without increasing the number of parameters, they could expand the local receptive field of the layer convolution kernel and improve the classification accuracy. This paper carried out the experiments on Caltech-10 1 database, Caltech-256 database and Uebingen animals with attributes database. The results show that the D-MobileNet models can achieve better classification accuracy than that of MobileN et, and can improve the classification accuracy by up to 2% . [ABSTRACT FROM AUTHOR]

Pedestrian detection is a task which is necessary among all tasks leveraged in automatic driving domain. Traditional pedestrian detection algorithms took fully advantage of image data, unable to obtain depth of objects. To address aforementioned issue, this paper proposed a method based on raw LiDAR point cloud data. The proposed method combined traditional moving object detection in LiDAR data and point cloud recognition by deep learning, and was capable of perception and pedestrian detection without images, obtaining 3D location of pedestrian, the refore helping central control system make reasonable decisions. This method experimented in 3D object detection task of KITTI dataset, obtained 33.37% AP (average precision) on mode rate cases. The results show that the proposed method performs better than other algorithm based on LiDAR data, which hence indicates the effectiveness of proposed method. [ABSTRACT FROM AUTHOR]

In order to ensure the load balance of network storage and to avoid unrecoverable loss in case of node or disk failures, this paper proposed a distributed storage and coding scheme based on balanced data placement strategy, which gave different solutions for large caches and small caches . Firstly, it extended the Maddah scheme to multi-server system, and stored each file as a unit in the data server by combining balanced data placement strategy to solve the large-scale cache problem. Then, it extended the interference cancellation scheme to the multi-server system. It used the interference cancellation scheme to reduce the peak rate of the cache, and proposed the linear combination of cache segments by combining balanced data placement strategy so as to solving the problem of small caching . Finally,it carried out simulation experiments in one and two parity check server systems respectively through Linux-based NS2 simulation software. The simulation results show that the proposed scheme can effectively reduce the peak transmission rate. It achieves better performance comparing with the other two new caching schemes . In addition, although distributed storage limits the ability to combine content from different servers into a single message, resulting in performance loss of coding and caching schemes, it can make full use of the inherent redundancy in distributed storage systems, thereby improving the performance of storage systems. [ABSTRACT FROM AUTHOR]

Aiming at database leakage caused by abnormal database user behavior, this paper proposed a database user anomaly detection method based on K-means and naive Bayes algorithm. Firstly, the K-means clustering method obtained users' grouping based on the user' s query statements and query results in the database historical audit logs; then, the naive Bayes classification algorithm constructed the user anomaly detection model. Compared with the model constructed by naive Bayes classification alone, the simplified representation of user behavior profile reduces computational redundancy and reduces training time by 81%. Applying K-means clustering method to obtaining users' grouping improves the detection accuracy by 7.06% and the F1 value by 3.33%. Experiments show that the proposed method greatly reduces the training time and achieves better detection results. [ABSTRACT FROM AUTHOR]

In view of the imprecision problem for PAD (pleasure,arousal,dominance) prediction, this paper proposed clustering PSO-LSSVM model combining LSSVM optimized by PSO and affective clustering analysis. Firstly, it selected three emotion speeches of TYUT2. 0 emotional speech database and Berlin voice library, and extracted emotion features. It established single emotional dimension PSO-LSSVM models for three single emotion and the mixed emotion dimension PSO-LSSVM model for three emotions based on emotion features and P,A and D values. The method used mixed emotion dimension PSO -LSSVM model to predict the P, A and D values of the test set, and calculated the distance between the predictive PAD and the PAD of the basic emotion. Finally it clustered the emotion whose distance was greater than the threshold into mixed emotion, and clustered the emotion whose distance was less than the threshold into the nearest emotions, then used the corresponding emotional dimension regression model to predict its P, A and D. The research shows that the predictive error of clustering PSO-LSSVM regression model to P, A and D is smaller than that of LSSVM and PSO-LSSVM model,and the correlation between the predicted value and the tagged value is stronger. So the clustering PSO-LSSVM regression model is more reliable and accurate in predicting P,A and D values. [ABSTRACT FROM AUTHOR]

Based on the data files provided by the MIT-BIH, this paper extracted the characteristic information of ECG signals by wavelet transform, and studied the classification and recognition of common signals. This paper mainly designed and implemented three classification algorithms based on softmax regression and neural network. Simulation experiments show that the training speed of a suitable neural ne twork algorithm is faster. With fewer iterations, the accuracy rate of classification recognition is more than 90%. [ABSTRACT FROM AUTHOR]

In the off-line signature, LBP of background pixels, as well as pixels inside the strokes, accounting for a large proportion in a signature bitmap, are nearly the same. Consequently, they have great interference to describe the characteristics of signature ' s strokes. So this paper proposed LBPC-based feature that mainly calculated the histogram of the local binary patterns on signature's contour and removed some useless patterns by adding a rule to improve the effectiveness and robustness of LBP. Besides, seen as an improvement for directional chain code for its limitations in the signature verification, this paper introduced LCPC-based feature which aimed at computing the statistical features of the local contour patterns co-occurrence. Then, it applied the PCA to two above combined features due to the huge dimensionality. Finally, to evaluate the performance of proposed method, using SVM classifier, it conducted experiments on MCYT and GPDS open databases and the achievable average error rates were 13. 51% and 12. 97% respectively. Moreover, comparisons with other methods on the same datasets provide evidence that the proposed method obtains lower average error rate than others. [ABSTRACT FROM AUTHOR]

Most recommendation systems based on the network structure suffer from lack of diversity, so that the paper proposed a recommendation system of big data based on multi-weight projection of bipartite network. Firstly, it abstracted the basic information of datasets, and applied items-users lists as an input to Levenshtein distance algorithm to compute similarity of each property. Then, it computed the number of common neighbors of the nodes in the bipartite network, the degree of common neighbors of the nodes in the bipartite network and degree of each node in bipartite network, and also computed the triple weights of each side of the bipartite network. Lastly, it adopted the enhanced bipartite projection technique to abstract the potential links of the bipartite network to realize the link prediction based on similarity. It realized the experiments based on both of big dataset and small dataset. The results show that the proposed algorithm outperforms different kinds of recommendation systems in terms of accuracy and coverage of recommendation, at the same time, it outperforms the other recommendation system based on network structure. [ABSTRACT FROM AUTHOR]

In order to reduce the resources occupied by the attribute-based encryption algorithm, this paper introduced an improved attribute-based outsourced encryption algorithm based on secure data access control attribute-based encryption. It outsourced the complex bilinear pairings in the encryption computing to the fog node to reduce the user's computation overhead. By simplifying the system parameters, it reduced the random number generated by attribute authority for every attribute to shorten the length of ciphertext and the key, as well as the storage cost of the system. And it gave the security proof of the scheme. [ABSTRACT FROM AUTHOR]

This paper proposed halftone self-hiding algorithms, and a secret image can be hidden in dot diffused or error diffused halftone images. Based on the data hiding algorithm by dual conjugate dot diffusion, it realized a dot diffused gray-scale halftone self-hiding algorithm, and extended this algorithm to an error diffused halftone image, and then proposed an error diffused gray-scale halftone self-hiding algorithm. Based on the data hiding algorithm by new color conjugate error diffusion and the data hiding algorithm by new color conjugate dot diffusion which were the latest color halftone visual cryptography, it proposed color halftone self-hiding algorithms in dot diffused and error diffused halftone images. It implemented experiments of four algorithms and compared the performance of different algorithms. The results show that the color halftone self-hiding algorithm based on dot diffused and error diffusion has the best performance under the same parameters. [ABSTRACT FROM AUTHOR]

In many applications of mobile crowd sensing, participants should collect continuous data over a period of time in this scenario on which existing research lacks consideration. For this scenario, this paper proposed a participant selection mechanism which was time window dependent. This mechanism included a participant selection method which was time window dependent based on dynamic programming algorithm. The target of the method was to maximize data benefits while cover time period of the task. It gave an updating mechanism of participant's reputation based on the willingness and data quality of the participant. Finally, simulation results show that compared with two common selection mechanisms, the proposed participant selection mechanism has better performance in terms of data reliability, data benefits and cost and has better prospects. [ABSTRACT FROM AUTHOR]

This paper proposed a construction framework of pet knowledge graph. It designed and constructed the schema layer in a top-down manner and constructed the data layer by extracting knowledge from semi-structured and unstructured data. For entity extraction of unstructured data, this paper proposed a symptom-named entity recognition method which combined conditional random field (CRF) and pet symptom dictionary. The method used symptom dictionary to identify the text and obtained the semantic category information, and then combined CRF and the semantic information to identify symptom-named entities. The experimental results show the effectiveness of the method. It selected the attribute graph model supported by the OrientDB database for knowledge representation. The knowledge graph used the OrientDB graph database for knowledge storage. In addition, it showed the constructed pet knowledge graph. [ABSTRACT FROM AUTHOR]

The traditional method of calculating the similarity of the Microblog users based on the user's blog content and the number of common friends has the problem of excessive potential error, and the similarity calculation model based on the user's multi-source background information has high computational complexity and ignores the user's interest and other issues. this paper put forward a method to calculate the comprehensive similarity combining user's interest and background information (BIBS). The method extracted the user's interest from the user's tag. When the user's tag was missing, it indirectly obtained the user's interest by clustering the important user's in the user's attention network, and calculated the user's interest similarity. Then it calculated the background similarity of the user according to the background information such as the gender, age and location of the user, so that it hierarchically mined the most similar users. Experiments and analysis based on the data of Sina Microblog show that compared with MISUR algorithm based on the similarity of multi-source information, the proposed method can improve the accuracy, recall rate and F-measure by 8.1%, 16.7% and 13.6% respectively with less time consuming, which proves the effectiveness and accuracy of the BIBS method . [ABSTRACT FROM AUTHOR]

Aiming at solving the problem of big data's large scale and complex computation, this paper adopted the idea of two-stage progressive clustering, and proposed a parallel computation algorithm for big data clustering based on MapReduce. In the first stage, this method acquired the initialized clustering center through Canopy algorithm, in order to find relatively accurate cluster center points quickly. In the second stage, it presented a novel scheme of parallel computation based on Map Reduce framework, which maked each data node cluster or merge around its adjacent Canopy center node. In this way, the algorithm could make the procedure of data clustering fast and accurately. The results of the experiments deployed on MapReduce show that this algorithm can effectively improve the efficiency of parallel computing, reduce computing time, and improve big data's clustering accuracy. [ABSTRACT FROM AUTHOR]

In traditional and relational database, it is difficult to meet the needs of data storage and quick retrieval due to such huge amount of data. For this problem, this paper studied the mapping model of data file field, the relationship between data file field and HBase column, and the importing execution plan, which had solved the heterogeneity of file object and the universality of storage transformation. It put forward the "RowKey" generation rules and algorithm, and provided the algorithm of data transformation based on mapping model and "RowKey". Then according to the mapping rules of data file fields with HBase table column, the data in data file was transformed into HBase. It realized the fast data access and retrieval according to prefix matching in "RowKey" or keyword matching, which had strong commonality and could be widely used in HBase large data storage applications. [ABSTRACT FROM AUTHOR]

With the development of deep learning, more and more deep learning models have been applied to the task of relation extraction, but traditional deep learning models can't solve long distance dependence problems. At the same time, distant supervision will inevitably generate wrong labels. For these two problems, this paper proposed a distant supervision relationship extraction method based on GRU (gated recurrent unit) and the attention mechanism. First, it adopted the GRU neural network to extract text features and solve long-distance dependence problems. Second, it constructed a sentence-level attention mechanism on entity pairs to reduce the weight of noise sentences. Finally, based on the real data set, by calculating the accuracy rate and recall rate, and drawing the PR curve to prove the proposed method has achieved significant progress compared with some existing methods. [ABSTRACT FROM AUTHOR]

Aimed at the WSN' s problems, such as limited node' s energy, low precision of measurement and short effective lifetime of network, etc., this paper proposed a data fusion algorithm based on abnormal data preprocessing and adaptive estimation, ADAEWF. In order to improve the algorithm' s reliability, this paper firstly proposed the data preprocessing mechanism based on abnormal data detection,simple majority principle and node's comprehensive support function. To reduce the measurement error' s effect on the fusion precision, this paper proposed an adaptive estimation weighted data fusion algorithm. Then, this paper established the WSN' s simulation model and obtained the mean square error and network 's effective lifetime under ADAEWF, AFWDF and arithmetic mean value algorithm, respectively. Simulation shows that, in terms of both fusion accuracy and network ' s effective lifetime, ADAEWF outperforms AFWDF and arithmetic mean value algorithm, and ADAEWF algorithm has superiority in improving the effectiveness of data fu sion, network lifetime and fusion accuracy. [ABSTRACT FROM AUTHOR]

This paper proposed a linear regression algorithm to integrate collaborative filtering based on the data sparse influence of the traditional collaborative filtering algorithm. Firstly, it built a similarity matrix between the user and the project based on the user’s rating of the project, as well as the user and the project’s own characteristics. Secondly, based on the similarity matrix, it selected the user and project nearest neighbor set. It predicted the score that the users had graded respectively by the way of collaborative filtering algorithms based on the user and the project. And it would take the difference between predicted scores and the real scores as features to generate new training data, and regarded the new training data as the input of the linear regression model. Finally, according to the training model, it could predict the unknown score, and used the top-N algorithm to generate the recommended list. It conducted the experiment on the MovieLens data set. The experimental result shows that the proposed accuracy of the new algorithm improves compared with the traditional collaborative filtering algorithm. [ABSTRACT FROM AUTHOR]

In order to solve the low efficiency and practical application of predicting the potential phone replacement user, this paper designed and built a phone replacement prediction system based on big data platform. This system firstly captured signaling data from multiple network interface and collected external data. Through the parse platform, the data from network interface would be distributed, decoded, synthesized and correlated, and the external data would be processed by ETL tools, and then storing processed data into IIDFS. Further, this paper established a phone replacement prediction model, which based on logistic regression, using spark components in the big data platform and output the potential phone replacement users. Finally, this paper chose part of the western city ’ s user data for system testing. The result shows that the prediction accuracy of the phone replacement prediction system is up to 71%. It can preferably recognize potential phone replacement users, and provide reliable support for the precise marketing of operators and mobile phone manufacturers. [ABSTRACT FROM AUTHOR]

In order to improve the robustness and efficiency of video retrieval for human activity, this paper proposed a fuzzy weighted method for human activity video retrieval. This method used 3D Harris operator to detect the spatio-temporal interest points in the video, and extracted the gradient information of these points to construct feature vector to describe video. Then it used fuzzy clustering method to construct fuzzy clustering feature vector, to improve the ability of anti-interference of feature vector. And then, it matched pair of gradient vector in the fuzzy clustering feature vectors to construct fuzzy weight matrix, and calculated the similarity between the query video and each video in the database. Finally, it carried out the video retrieval experiment on the KTH database, and carried the evaluation out with three metrics of accuracy, recall and retrieval time, which proves that the performance of this method is the best. [ABSTRACT FROM AUTHOR]

In order to meet the needs of the network under the background of big data, and eliminate inferior data interference data knowledge high precision requirements of large data transmission, this paper proposed variable size adjustment scheme based on the algorithm to expand the network of large data knowledge opportunistic scheduling. Based on the analysis of large data network characteristics, it normalized the adaptive vector encoding, capture the heterogeneous characteristics of large data network, using multi order back-propagation network of heterogeneous data, and then through the real-time transmission of large data network to achieve opportunistic scheduling. At the same time, the knowledge engineering system composed of network data segmentation of fine-grained big data based on the multidimensional feature dimension, the granularity of knowledge transformation was known, then adjusted the size of the dynamic characteristics, making big data set of knowledge engineering with linear characteristics and clear geometric characteristics, improved the accuracy of knowledge acquisition through knowledge expansion. The experimental results are compared with the algorithm based on fine grained knowledge acquisition, which proves the high reliability, real time and high efficiency of network data transmission. [ABSTRACT FROM AUTHOR]

In order to solve the problem of inefficient query in relational database keyword search scheme based on schema graph, this paper proposed combined candidate network evaluation( CCNE) method. This method designed a new combined candidate network(CCN) structure, which could effectively avoid the redundant operation caused by the repeated structure between candidate networks(CN) in the traditional method. At the same time, it proposed an improved CN generation strategy to improve the efficiency by avoiding generating redundant CN and narrowing the traversal range. Finally, based on CCN, this paper designed a CCN evaluation algorithm, which greatly reduced the complex database query operations required by traditional methods and further improved the query efficiency. Experiment based on real data sets show that CCNE method can improve query efficiency without any loss of query results. [ABSTRACT FROM AUTHOR]