According to the characteristics of pipeline structure and multi-core processor structure for packet processing in network detection applications, the horizontal-based parallel architecture model and tree-based parallel architecture model are proposed for packet processing of Snort application. The principle of a tree-based parallel architecture model is to use pipelining and flow-pinning technology to design a processor that is specifically used to capture data packets, and other processors are responsible for other stages of parallel processing of the data packets. The experimental comparison and analysis show that the tree-based parallel architecture model has higher performance on the second-level cache hit ratio, throughput, CPU utilization, and inter-core load balancing compared to the horizontal-based parallel architecture model for packet processing of Snort application.

It is well known that imbalanced datasets are a common phenomenon and will reduce the accuracy of classification. For solving the class imbalance problem, this paper proposed the mixed weighted KNN algorithm. According to the imbalance between the classes, this algorithm assigns each sample of datasets an inverse proportion weight, and then it combines with the distance weight, making the weight of the training sample close to the test sample greater. In order to improve the operating efficiency and make it easy to handle massive datasets, we implemented the parallelism of MW-KNN based on the Hadoop framework. Experimental results show that the proposed algorithm is simple and effective.

The recommendation system is a new technology used to recommend products for customers from huge amounts of products by inferring objective users’ preferences based on their personal information or online behavior. This paper studied the main personalized recommendation technology for current e-commerce. It proposed a hybrid recommendation algorithm based on opinion mining. This system combines web data mining technology, i.e., takes advantage of user-generated content by mining customers’ online reviews. It is well known that online reviews can directly reflect a customer’s real emotions and expectations, so it is appropriate to extract a customer’s latent interest and preference from his/her reviews, thus refining recommendations and improving accuracy. Meanwhile, an experiment was conducted and the result demonstrated that our system could generate a reliable and realistic recommendation.

Image representation is a key step in image advertising recommendations. Traditional image representation methods, based on the local description, generate a histogram of visual words to represent images. However, it is very difficult to establish a discriminative and descriptive codebook with the local description only. Therefore, we propose a novel image representation method by integrating visual saliency, color feature and local description. Moreover, the proposed multi-cues image representation has been applied to a new image advertising scenario, i.e., delivering image advertisements in a list of images, such as the results of an image search. To evaluate our proposal, we have crawled a dataset, named Pop2016, which consists of image lists and advertising images with 31 pop labels. The performance of the advertising recommendations is measured in terms of the precision@n and the mean average precision. Experimental results show that the proposed algorithm outperforms several traditional methods.

A semantic query expansion method is put forward based on the comprehensive weighted algorithm of semantic similarity. We combine the ontology-based query expansion and corpus-based query expansion. If the query term matches the concept, we calculate the similarity between concepts, construct the connected graph of correlation among the ontology concepts, and expand the semantic query according to the threshold value. Otherwise, we adopt the Naive Bayes algorithm to calculate the co-occurrence probability between the word set and concepts as the relevancy of semantic query expansion. The experimental results show that this method can improve the retrieval performance effectively, with the Pr@30 index being improved by 41.97% compared to the traditional non-extensible query method.

In recent years, with the increase of the resource exploitation, mining depth is getting deeper and deeper. Meanwhile, the lives of mining workers have been threatened strongly. In order to reduce the probability of rockburst, researchers have carried out in-depth research on rockburst. He Manchao, the academician of the State Key Laboratory for GeoMechanics and Deep Underground Engineering, has initially simulated the occurrence of rockburst in the laboratory as well as studied the mechanism of rockburst. Because the amount of data accumulated in the experiment is as large as 1000T, using these valuable experimental data becomes a difficult problem. Therefore, we have introduced big data technology into the field of rockburst. We have designed and realized the automated collaborative analysis system of the rockburst mechanism based on big data. We have used acoustic emission sensors as data collection methods and selected the multi-task online learning algorithm for data processing and analyzing. We have achieved the selection of the inflection point in the process of force changes using Matlab. In addition, the inflection point to be checked in the system can obtain the corresponding part of analysis diagrams. The theoretical analyses and experimental studies show that the automated collaborative analysis system can have an obvious influence on rockburst data processing, which provides a good foundation for the study of the rockburst mechanism.

After studying the topological structure of neighboring nodes in the WSN, this paper presents a local Combination Location (LCL) algorithm by combining principal manifold learning and the nonlinear dimension algorithm. This algorithm is particularly suitable for determining the relative locations of sensor nodes in large-scale, low-density WSNs, where the low connectivity between nodes and the large ranging error between long-distance nodes usually make accurate localization quite difficult. In this algorithm, based on the pair-wise distance between each node and its neighbour nodes within a certain communication range, the local geometry of the global structure is firstly obtained by constructing a local subspace for each node, and those subspaces are then aligned to give the internal global coordinates of all nodes. Combined with the global structure and the anchor node information, we can finally calculate the absolute coordinates of all unknown nodes by the least squares algorithm.

A playfield detection method exploiting both color and local consistency features are proposed. Color feature is used in existing playfield detection, which does not effectively remove green pixels that do not belong in the playfield. To solve this problem, local consistency feature is introduced, and the playfield is detected using both color feature and local consistency feature. To determine the detection threshold of local consistency, a two-dimensional histogram based method and a color constrained Otsu (cOtsu) based method are proposed, which are based on the principle of color characteristic and local entropy characteristic of playfield pixels, respectively. Experiments show that the proposed method is more effective and is able to detect playfield in several typical environments.

In studies related to the construction of domain ontology, the acquisition of concept meaning has not received enough attention. According to the definition of the concept, the acquisition of concept meaning is a requisite task in the process of ontology construction. This paper studied the automatic acquisition of concept meaning based on HowNet and researched the problem of meaning acquisition for complex terms and synonym removal. Grounded on concept meaning, this paper put forward a sememe suffix tree algorithm and applied it to the construction of ontology taxonomy. Compared to traditional algorithms, this method is more efficient and comprehensible. This paper implemented the methods to the domain of ethnic minorities, and the experimental results showed that this paper is referable.

Software defect prediction (SDP) models facilitate software practitioners to find out defect-prone software modules in software. Software practitioners can then test these defect-prone software modules with limited testing resources to minimize software defects. Among various SDP models, Naive Bayes (NB) has been widely used in SDP because of its simplicity, effectiveness and robustness. The NB classifier is an effective classification approach, especially for data sets with discrete attributes. In NB, the attributes are assumed to be independent and thus equally important. However, in common practice, the attributes of software defect data sets are usually continuous or numeric, and because they are designed for different purposes, their contributions to prediction are different. Therefore, this paper proposes a new NB method called TSWNB, which contains two stages: feature (i.e. attribute) discretization and feature weighting. More specifically, for the stage of feature discretization, we make the comparison between two discretization methods i.e. equal-width discretization method and equal-frequency discretization method, and identify the most appropriate one. For the stage of feature weighting, we use the feature weighting technique to alleviate the equal importance assumption, which combines the obtained feature weights into the NB formula and its likelihood estimations. To evaluate the proposed method, we carry out experiments on 5 software defect data sets of NASA MDP provided by PROMISE repository. Three well-known classification algorithms and two feature weighting techniques are included for comparison. The experimental results reveal the effectiveness and practicability of the two-stage feature weighting method TSWNB.

Integer defects are an important cause of software quality degradation. An explicit expected output plays an important role in the traditional theory of software testing, but it is very difficult for much software to get the expected output since ascertaining the validity of the actual output is very hard. Integer bugs are always ignored because of the Test Oracle problem. A metamorphic relationship that can find out the potential error is presented. The experimental results show that the mean of integer bugs detection based on the metamorphosis relation can detect the invisible unexpected output, which is unable to get in traditional means. In addition, the effectiveness of detecting integer defects is improved.

In network systems, a job with a large amount of work-requirement is usually divided into multiple tasks to achieving parallel computing. However, if any task is failed due to random task failures or server failures, the entire job cannot be complete. In traditional redundant computing, copies of tasks are initiated whenever the tasks are found to be failed. This is not an efficient approach from the perspective of the performance. Real-time parallel computing is more high-efficient, which makes tasks and their copies run simultaneously. In this paper, to evaluating the reliability of such a job, we first describe a complicated parallel and redundant computing environment as multiple minimal-job spanning trees (MJST) consisting of tasks and servers. Then, we design two operators of MJSTs to systemically analyze complicated failure correlations among multiple MJSTs. Finally, an algorithm based on the Bayesian theorem is presented to evaluate the reliability of a parallel job with real-time parallel computing. Illustrative examples are provided.

Refactoring is a very important technology to improve the reusability and maintainability of existing code, and it is widely used in software development. In order to extract the code segment into a new method easily and cover the shortage of Eclipse in refactoring, the method of Context-Aware Automatic Code Segment Extraction and Refactoring (CAACSER) is proposed. By analyzing the context of the code, the input parameter class, and the output parameter class are introduced to handle complex code segments. The experimental results show that the CAACSER effectively solves some problems and drawbacks of many existing tools in code segment extraction, which acts as a basic step for realizing automatic and semi-automatic refactoring methods. The visualization tool of CAACSER can also carry out reasonable optimizations of the code without changing the systems’ behaviors.

The use of point cloud datasets is an inevitable trend in the analysis of natural scenes. In this paper, we propose a semantic segmentation network architecture that consumes 3D point clouds directly, which can efficiently avoid mapping 3D point clouds to 2D images. Experimental results indicate strong performance that is on par with or even better than state-of-the-art methods for semantic segmentation on the Stanford semantic parsing dataset.

This paper proposes a novel image restoration algorithm to eliminate light attenuation in underwater environments. The proposed algorithm employs the distance-dependent formation to model the degradation process where light travels in underwater. We use a homomorphic filter to get rid of the non-linear in the distance-dependent model. The restoration underwater image is then obtained by solving a Poisson equation that is derived based on the similar distance of neighbourhood pixels. The experiments show that the restoration image reveals improved contrast and clear details.

Spontaneous imbibition has been well investigated due to its impacts on oil recovery in conventional and fractured reservoirs. Differentiated in relation to the conventional reservoir, the tight reservoir is characterized by its ultra-low porosity and permeability. The scope of current research is to study the mechanism of imbibition under conditions of tight cores and surfactant solution. A series of experiments was conducted to determine the effects of tight core imbibition including three kinds of common surfactant solutions. The main factors of low IFT (interfacial tension) and wettability alteration are examined. The experimental results indicate that the original wettability affects the imbibition oil recovery, since low IFT and wettability alteration constitute beneficial parameters for imbibition as it concerns adhesive work reduction. The parameter of N_Bm^-1 was not applied for optimal IFT or CA (contact angle) calculation concerning porous media with reduced capillary scale. Future research concerning the imbibition mechanism should take the property of surfactant into account.

An automatic fault diagnosis method for the wind turbine generator system (WTGS) driven by vibration signal is proposed in this paper. In this method, the vibration signal is used to drive the notch filter network directly, and the frequency selection characteristics of the notch filter are used to extract the fault feature frequency of WTGS components. Then, the extracted fault feature frequency is encoded and a neural network classifier is used to achieve the automatic fault diagnosis of WTGS. In addition, the vibration intensity is calculated to evaluate the fault degree of the WTGS. The innovation of this paper is that the fault feature frequency of the WTGS is derived from parameters of the notch filter rather than the vibration signal itself. The practical on-site application shows the effectiveness of the proposed method, which is of great significance for improving the efficiency of fault diagnosis of WTGS and realizing the batch diagnosis of the fault of WTGS.

Focusing on the complex constellation of the BeiDou satellite navigation positioning system in China, this article analyzes the relationship between zero baseline and ultrashort baseline according to the residual value of carrier observation, signal to noise ratios, and elevation angles of different constellation satellites. A stochastic model based on the distance between the satellite and Earth, signal-to-noise ratio and elevation angle was proposed, which applied to the baseline solution of the Beidou triple-frequency single epoch. Analysis of measured data shows that compared with the traditional sinusoidal trigonometric function model, although the proposed model has no obvious effect on improving the positioning accuracy, the success rate of the ambiguity resolution of 4.2m, 4.17km and 8.84km are effectively improved by 0.1%, 4.7% and 1.1%, which enhanced the stability of calculation results.

With the rapid development of the e-commerce economy in urban and rural areas, China's logistics industry has entered a stage of transformation and upgrade. First, this paper introduces the Supply-Chain Operations Reference-model as a theoretical reference for index selection. Then, after comparing the BP neural network and linear regression analysis, we chose the BP neural network analysis method, which is more stable and accurate in forecasting e-logistics demand scale in urban and rural areas. Finally, according to the results of the data analysis, this paper divides the development of e-logistics demand in urban and rural areas into two stages and discusses the reasons for the formation of these two stages in detail. This job not only provides a new perspective for the study of rural e-commerce and urban and rural e-logistics demand prediction, but also provides a theoretical reference for the formulation of government policies and farmers’ participation in rural e-commerce.

Based on summarising and analyzing the research status of policy research and the technology of concept description, this paper proposes an effective mining method of recessive lineage relationship between policies. This method introduces the theory of factor space and gives a method for decomposing concept factor. We introduce the recessive gene as a new "synonym" and calculate the fitting degree of policy texts to reveal the recessive lineage relationship, which cannot be reflected through the direct calculation of similarity. Finally, we use the articles of law as the set of policy texts to carry out a large number of experiments. The comparison and analysis of the experimental results verify the effectiveness of the proposed method.

In this paper, we propose a comparative summary generation method and conduct key technologies research. We collect prior knowledge from the Internet via a light knowledge structure, extract core information from original literature, divide subtopics of two major topics with AGNES clustering to get the common and independent subtopics, and get comparative information with subtopics alignment and property alignment. We test the performance of each module to prove the validity of the proposed methods. Finally, we design and develop a comparative summary generation system, and the application in the nursing field shows that it can present users with useful information to facilitate the scientific research process.

In view of data sparsity and long-range dependence when training the N-Gram Mongolian language model, a Mongolian Language Model based on Recurrent Neural Networks (MLMRNN) is proposed. The Mongolian classified word vector is designed and used as the input word vector of MLMRNN in the pre-training phase, and the Skip-Gram word vector with context information is used at the input layer so that the input contains not only semantic information, but also rich context information. It effectively avoids the problem of data sparsity and long-range dependence. Finally, the training algorithm of MLMRNN is designed and the perplexity is used as the evaluation index of the language model to test the perplexity of N-Gram, RNNLM and MLMRNN on the training set and test set, respectively. The experimental results show that the perplexity of using MLMRNN is lower than that of other language models, and the performance of the language model is improved.

Dynamic network research has been a new trend in recent years. Based on the influence of vertex behavior on community structure, this paper studies signed network dynamic community mining. Firstly, the set pair connection degree is introduced to describe the relation between vertices, and the edge prediction model of signed network is proposed by taking into account the variability of the relation between vertices. Secondly, based on the prediction model, a set pair signed networks dynamic model is proposed by adding time axis T to the signed network. Then, based on the dynamic model, the evolution of signed networks and community discovering are studied. Finally, network evolution law and community stability are analyzed by using the connection trend and connection entropy in set pair theory, and the accuracy and validity of the dynamic community mining algorithm are verified by experiments.

With more and more cameras installed in public places, video surveillance systems play an increasingly important role in public safety. Research on intelligent video monitoring, especially activity recognition, is attracting increasing attention in the field of image processing. Unlike activity recognition of a single tracking object, group activity is more complex and difficult to recognize. To design a fast real-time group activity recognition algorithm without other auxiliary data, low computational cost is our focus. There are four steps for our group activity recognition system: preprocessing the captured videos, extracting foregrounds from backgrounds, tracking multiple objects and recognizing group activity. To remove noise in each frame image, the combination of the Gaussian filter algorithm and median filter algorithm is used in the preprocessing step. Then, the Gaussian mixture model is adopted to extract the foreground image. To ensure low computational cost, real-time Cam-Shift is chosen to track group activity with morphological operations in the tracking step. In the recognition step, the changing histogram rate is defined as the measure of identifying group behavior. Here, the changing histogram rate refers to the number of changing histograms and changing proportions. Experimental results show that the group activity recognition algorithm proposed in this paper is effective with low computational cost.

The internal pressure and external incentive of construction area contributed to the study of supply chain management, construction calls for the construction supply chain management’s (CSC) research and practice. Combining with features of Data Envelope Analysis (DEA) and CSC, the basic framework of CSC comprehensive performance evaluation was built from the aspect of consumption and output. Under such framework, the indicator system for comprehensive evaluation was also constructed. Because traditional methods of DEA do not consider the impact of preferences of the input and output indicators for decision makers, an improved preference restraint cone DEA model was presented; an example was given for analysis. The results show that the DEA with the preferred model omissions the pseudo-effective construction supply chain and reflects the subjective preferences of decision makers. It is an improvement from past research of the model.

SOC is an important parameter of power batteries of electric vehicles. Its accurate estimation is vital to the correct implementation of the control strategy of the whole vehicle. It is strait to estimate SOC of the battery accurately using existing estimation methods. Aiming at the shortcomings in these methods, we proposed to establish an estimation model for battery SOC using principal component analysis (PCA) algorithm in this study. However, unable to extract non-linear factors in parameters, PCA algorithm would bring about an estimation error of battery SOC; thus, we proposed to establish an estimation model for battery SOC using kernel principal component analysis (KPCA) algorithm. The model was simulated and verified through experiments. After simulation, it shows that the improved model may adapt to a more complicated environment, meet the requirements of promptness and reliability, and has higher estimation accuracy with an average estimation error of 1.46%, which is better than that of Ah measurement method.

Knowing which types of functionality that human membrane proteins belong to is very helpful for understanding their functions. However, most existing online prediction methods have some disadvantages, including: 1) they obtain very low prediction accuracy, and 2) they can only predict single-functional classes of cytomembrane proteins in humans. To overcome the drawbacks, a new multi-label predictor, namely mMem-Hum, is proposed. In addition to predicting types of single-function membrane proteins, it can also predict multi-functional types. Specifically, discriminative features of membrane proteins are generated by using amino acid sequence information and evolutionary information, and then they are classified by a new multi-label classifier that utilizes label correlations. Experimental results reveal that the performance of mMem-Hum is significantly better than other existing forecasting methods. This indicates that mMem-Hum may become a promising prediction tool for classifying functional classes of cytomembrane proteins in humans.

The locations of apoptosis proteins in the cell determine their biological functions. So firstly, it is necessary to identify the subcellular locations of these proteins. In recent years, researchers have proposed a large number of prediction methods, specifically for apoptosis proteins. However, the vast majority of the methods have the following problems: (1) they utilize sequence-based methods rather than annotation-based methods for feature representation; (2) they ignore the negative impact of the imbalanced training dataset. In the work, a balanced predictor, GOIL-Apo, is proposed for dealing with the issues, which yields balanced solutions for predicting locations of apoptosis proteins. Firstly, by using gene ontology (GO) based methods, apoptosis proteins are represented as GO feature vectors. Subsequently, an ensemble classifier that fuses multiple SVMs with random under-sampling is proposed to deal with the data imbalance problem. Rigorous cross-validations show that the accuracy of GOIL-Apo is much better than the up-to-date predictors.

Artificial Intelligence (AI) has been developed to be robust on computing. Learning can be achieved by connecting to heterogeneous data using AI algorithms, such as the Artificial Neural Network. Knowledge can be learned, and rules in the database can be discovered by machines through heuristic algorithms. However, creativity has not been achieved by computers like the human brain by using AI algorithms individually. This research serves to explore a method to achieve creative solution generation by utilizing a relationship between intelligence and creativity, assuming intelligence is the subset of creativity. Under this relationship, the computing can be fulfilled using AI algorithms. The theories of achieving creativity is the guidance of this method.