Earticle

The remote sensing images are massively stored, so it is difficult for the traditional single-node mode to meet the real-time requirement for remote sensing image retrieval. In order to improve remote sensing image retrieval efficiency and accuracy, a kind of feature information MapReduce based remote sensing image retrieval algorithm is proposed in this article. Specifically, the color features and the texture features of the remote sensing image are firstly extracted, and then Map function is adopted to calculate the similarity between the remote sensing image to be retrieved and the image in the feature library according to the color features and the texture features, and finally Reduce function is adopted to collect the intermediate results of various node tasks and the remote sensing images are ranked by a descending order according to the similarity in order to obtain the remote sensing image retrieval result. The test result shows that the proposed algorithm can rapidly and accurately retrieve the remote sensing image, thus not only improving the remote sensing image retrieval efficiency, but also improving the remote sensing image retrieval accuracy.

Data Stream is a continuous set of data records. When data arrive at a very high speed and continuously, so predicting the class in timely manner is important. Class prediction of data Stream is an important task in data mining. Nowadays Ensemble Modeling technique growing rapidly in Data Stream Classification. Ensemble learning become popular because of its advantage to handle large quantity of data stream, means it can handle the data in a bulk and also it can handle concept drifting. Earlier studies, mostly focused on accuracy of ensemble model, prediction efficiency has not considered much because existing ensemble model predicts in linear time, which is enough for general or small applications and existing models works on integrating small number of classifier. But real world application have large volume of data stream so we need more base classifier to identify different patterns and build a high grade ensemble model. To overcome these challenge we propose height balanced tree indexing structure (Ensemble tree) of base classifier for fast prediction on data streams by ensemble modeling technique. Ensemble Tree handles ensembles as spatial databases and it make use of an R-tree like structure to achieve sub linear time complexity.

Spatial Cluster analysis is another important technique in the field of spatial data mining, especially the K-Means spatial clustering method, which can deal with spatial objects with geographical location and attribute. However, with the development of the information society, the spatial data grows explosively, but the serial algorithm has low computing efficiency and is difficult to process massive spatial data. Aiming at spatial with a double meaning of location and attribute, the paper designed and implemented K-Means spatial clustering parallel algorithm on Hadoop. Using Yahoo Weibo user data is to do clustering analysis. Finally, the visualization of clustering results was implemented by Google Map.

The percent of examinees’ first applications for a university can reflect the scientific research level of this university, so various universities start to research how to improve the percent of examinees’ first applications under the condition of not influencing the enrollment quality. For such research, C4.5 decision tree algorithm is applied to the postgraduate enrollment of a certain university. Specifically, examinees’ information is processed to select decision attributes and establish the decision tree so as to obtain the relation among examinees’ first applications, native place information, total points of initial examination and category of graduation universities from the rules extracted thereby. The mining result shows that this algorithm can correctly classify the graduation universities and assist the enrolling personnel to more effectively stipulate the enrollment guide for the targeted enrollment propaganda, thus to improve the percent of examinees’ first applications.

Opinions, the key influencer of human behavior and activity is ranked as of one of the strong factors that determine the effectiveness of one’s strategy and approach in terms of influential power and trend setting capabilities. This highlights the importance of sentiment analysis done upon the extracted data. Today, statistics have shown significantly that most opinions can be obtained via many social media platforms. Social media has provided a convenient platform for web users to comfortably share their thoughts and to boldly voice up. Having to process such huge amount of data, it is proposed that automated sentiment analysis is done when extracting social media data. Using an effective algorithm which produces meaningful information from raw data, the possibilities of venturing deeper into areas like decision making and influential thinking are simply limitless.

Collaborative filtering algorithm is the most used items recommendation algorithm. We find the k neighbors with the highest similarity by calculating user similarity and recommend items for users by the score of the neighbors of the items. In the paper, we propose a hybrid recommendation algorithm based on user similarity and attribute weights to solve user ratings sparsity. We obtained the weights of users like properties through learning user ratings records and combined with the user similarity for users to recommend item. Finally, we transplant the algorithm to HADOOP platform. Through the experiment, the improved collaborative filtering algorithm is better than the original algorithm in precision and parallel attribute.

Machine learning is ideal for exploiting the opportunities hidden in big data. It delivers on the promise of extracting value from big and disparate data sources with far less reliance on human direction. It is data driven and runs at machine scale. It is well suited to the complexity of dealing with disparate data sources and the huge variety of variables and amounts of data involved. And unlike traditional analysis, machine learning thrives on growing datasets. The more data fed into a machine learning system, the more it can learn and apply the results to higher quality insights. In this paper we propose a robust machine learning approach for dealing with large data set. Through experimental results, proposed method performs well on large data sets.

Support vector machine (SVM) has been widely applied to small-sample, non-linear and high-dimensional classifications. Many modified SVM algorithms were put forward in recent years. Some of them focus on SVM feature selection and some focus on SVM classification effectiveness. As different input vectors have significant influence on learning effect of decision boundary, this paper proposes a weighted multi-class support vector machine (WSVM) algorithm. The algorithm gives different weights to features according to the importance of their information. WSVM algorithm establishes decision boundaries based on weights and is used to classify educational resources. Experimental results indicate that the method achieves relatively good classification effectiveness.

This paper proposes an improved clustering algorithm on the basis of the characteristics of sampling and density. The initial k value and initial center are determined by sampling and density, and parallel improvement is based on the HADOOP platform. Through the experiment, the improved K-Means algorithm has good parallelism.

This paper focuses on clustering the lines of Shakespeare Sonnets. Sonnet Line Clustering (SLC) is the task of grouping a set of lines in such a way that lines in the same cluster are more similar to each other than to those in other clusters. K-Means clustering is a very effective clustering technique well known for its observed speed and its simplicity. Its aim is to find the best division of N lines into K groups (clusters), so that the total distance between the groups’s members and corresponding centroid, is minimized. A new algorithm Sonnet Line Clustering with Random Feature Selection SLCRFS is proposed. To validate the process external validation or internal validation is done. Since, internal validation has no considerable impact in conducting research this work concentrates on the measures of external validation. Entropy and Purity are frequently used external measures of validation for K-Means. The proposed approach uses entropy as performance measure. The clusters formed are evaluated and interpreted according to the Euclidean measure between data points and cluster centers of each cluster. This paper concludes with an analysis of the results of using the proposed measure to display the clustered sonnets using K-Means algorithm with minimum entropy for different feature sets.

Heterogeneous networks, composed of multiple types of objects and relationships, are ubiquitous in real life. Although many methods have been proposed for community detection in homogeneous networks which contain only one type of objects and one type of relationships between these objects, effective direct clustering objects of all types in heterogeneous networks without heterogeneous-to-homogeneous transformation remains a challenge. To achieve this goal, we propose a three-phase method for clustering star-structured heterogeneous data based on diffusion path. By adopting the principle that central objects are more important than attribute objects, we firstly assess the proximity of central objects in terms of their connected objects of all types, then based on which we cluster central objects, and thirdly we detect attribute objects groups according to their associated central objects. Finally, experiments on a real-world data set show the effectiveness and efficiency of the proposed methods.

Educational data mining has received considerable attention in the last few years. Many data mining techniques are proposed to extract the hidden knowledge from educational data. The extracted knowledge helps the institutions to improve their teaching methods and learning process. All these improvements lead to enhance the performance of the students and the overall educational outputs. In this paper, we propose a new student’s performance prediction model based on data mining techniques with new data attributes/features, which are called student’s behavioral features. These type of features are related to the learner’s interactivity with the e-learning management system. The performance of student’s predictive model is evaluated by set of classifiers, namely; Artificial Neural Network, Naïve Bayesian and Decision tree. In addition, we applied ensemble methods to improve the performance of these classifiers. We used Bagging, Boosting and Random Forest (RF), which are the common ensemble methods used in the literature. The obtained results reveal that there is a strong relationship between learner’s behaviors and their academic achievement. The accuracy of the proposed model using behavioral features achieved up to 22.1% improvement comparing to the results when removing such features and it achieved up to 25.8% accuracy improvement using ensemble methods. By testing the model using newcomer students, the achieved accuracy is more than 80%. This result proves the reliability of the proposed model.

The manifold-ranking based method is widely used in semi-supervised learning, and its performance is closely related to the structure of the constructed graph. In this paper, we propose a novel graph structure named natural neighbor graph and an algorithm to construct it. We apply the new graph structure into the framework of manifold-ranking based image retrieval. The greatest superiority over k-NN based method is that the free parameter k need not to be explicitly specified any more. We have shown that the manifold ranking algorithm based on our proposed graph structure performs better than k-NN graph. Experiments demonstrate that the proposed algorithm outperforms the state-of-the-art algorithms.

The spatial data set has much useful information, but the amount of volume is massive and the type is complex. It makes hard to analyze the spatial data. There are software tools for general data. Hadoop is one of the tools to process the big data. Hadoop can be used to analyze the large amount of spatial data. This paper proposed a data analysis technique for massive spatial data using Hadoop. We extend the grid based clustering algorithm to use Hadoop. The grid based clustering algorithm makes clusters with cells. Each cell has a number that counts contained objects. Only the cells who had the sufficient population can be join in clusters. The other cells ignored as noise. This paper proposed to enhance performance using Hadoop. In order to evaluate the enhancement of performance, the execution time is measured and compared. As the result, the proposed algorithm is 1.8 times faster than the original grid based clustering algorithm.

The radial basis function neural network (RBFNN) is a great potential artificial intelligence technology and can effectively realize the fault diagnosis for small sample and nonlinear problem. But the parameters of RBFNN model seriously affects the generalization ability and diagnosis accuracy on the great extent. So an improved differential evolution algorithm based on dynamic adaptive adjustment strategy is proposed to optimize the parameters of RBFNN model for obtaining the optimal RBFNN (DASDERBFNN) method. Then the proposed DASDERBFNN method is used to construct a new fault diagnosis (DSDRBFNFD) method. In the DSDRBFNFD method, the dynamic adaptive adjustment strategy is used to adaptively adjust the crossover probability (CR ) value according to the fitness value of current individual in the population for obtaining the improved DE(DASDE) algorithm. Then the selection of parameters in the RBFNN is regarded as a combination optimization of parameters in order to establish the objective function of combination optimization. The DASDE algorithm is used to search for the optimal value of objective function to obtain the better parameter optimization of the RBFNN (DASDERBFNN), which is applied in the fault diagnosis for constructing a new fault diagnosis (DSDRBFNFD) method. Finally, the proposed DSDRBFNFD method is used to diagnose the fault of the cylinder of the engine in order to validate the diagnosis effectiveness of the DSDRBFNFD method. The experiment results show that the proposed DSDRBFNFD method can obtain the higher accuracy of fault diagnosis and is effective fault diagnosis for the engine.

For the representation and reasoning of fuzzy knowledge for domain ontology, the paper presents the reasoning method based on fuzzy ontology. On a basis of the study of civil aviation emergency domain ontology, it gives fuzzy extension model of domain ontology based on fuzzy description logic from the perspectives of fuzzy modifiers and fuzzy concrete domains, and designs fuzzy rules by introducing weight concept and gives the representation and construction process of fuzzy rules on a basis of f-SWRL in connection with the inference application of situation analysis for civil aviation events. The experimental results show that the fuzzy extension method of domain ontology can make up for the issue that the existing domain ontology is inadequate in terms of fuzzy knowledge representation, and provide a good methodological support for making domain ontology perfect and inference application with the reasoning implementation based on fuzzy rules.

Text data classification and retrieval is an important field of artificial intelligence, but because of the different characteristics and different language syntax, classification and retrieval also will be different; classification and identification retrieval traditional text data, using training data and segmentation the algorithm does not consider the specific locale, it is often an error. To solve this problem, we propose a weighted feature-based classification method, according to this method, the text data can be quickly and accurately classify; experiments show that the proposed algorithm can effectively improve the accuracy and speed of classification and retrieval.

Cloud computing is regarded as a new computing mode in recent years, and has been widely applied. Its task scheduling affects the performance of the cloud computing system directly, and people pay more and more attention to the security problems of cloud computing. The paper introduces the trust in scheduling algorithm, improves and fuses the PSO and SA in order to make them complementary. By applying that to the task scheduling strategy of cloud computing, we can get a higher scheduling efficiency. We implement the proposed algorithm and verify its high efficiency through the simulation platform (CloudSim).

Prediction of network public opinion is a complicated prediction featuring poor information, small samples and uncertainty. A prediction model of network public opinion based on grey support vector machine (SVM) is specified to increase prediction accuracy. First, network data are preprocessed by text clustering, hotpot extraction and data aggregation. Then a time series model GM(1,1) is established and SVM is used to modify prediction outcomes of GM(1,1). At last, simulation experiment is conducted to test performance of the model. Simulation results indicate that grey SVM improves the prediction accuracy of network public opinion compared with traditional prediction models. The predictions have certain practical values.

The rise of globalization and market liberalization are changing the face of market competitiveness significantly. The appearance of modern technology in business processes has intensified the competition and put forth new challenges for service providing companies. To cope up with changing scenarios, companies are shifting their attention on retaining the existing customers rather hiring new ones. This is more cost effective and requires lesser resource as well. The phenomenon of abandoning the company by a customer is known as churn and in this context, anticipating the customer's intention to churn is called churn prediction. Data Mining and machine learning techniques, as applied to customer behavior and usage information, can assist the churn management processes. This paper used customer usage and related information from a telecom service provider to analyze churn in telecom industry. The decision trees and its ensembles, Random Forest and Gradient Boosted trees are used as underlying statistical machine learning models for building the binary churn classifier. The implementation part has been done using apache spark which is state of the art unified data analysis framework for machine learning and data mining. In order to achieve better and efficient results, the grid based hyper-parameter optimization is applied.

High utility itemset mining algorithms should weigh itemset statistical information with their semantic importance to generate a more accurate and sensible description of the itemset utility. A novel High Utility Itemset tree (HUI-tree) structure, which is an extended prefix-tree structure for the storage of compressed utility information about itemsets, is proposed to address this issue. Moreover, FHUI-Growth, a fast approach for high utility itemset mining algorithm is developed for mining high utility itemsets. Mining efficiency is achieved with three new techniques: (1) both frequency statistic and complex itemset utility information can be compressed into the condensed HUI-tree structure, which successfully avoids multiple database rescan, (2) a part of the utility calculation process can be simplified because a tighter bottom bound pruning constrain can be obtained through the HUI-Tree, and (3) the costly tree scan operation is converted into the item conditional projection matrix row and column computation, which effectively reduces the mining process. Evaluations of the testing data set show that the execution performance and scalability are better than the classical Two-Phase algorithm.

With the rapid development of information technology and the Internet, people entered the era of information overload. Recommended system is an effective tool to solve the problem of information overload, it is based on the historical behavior of users and other records of interest to the user modeling, and then use the model to create user interest personalized recommendation, the interested user information, products. Online Intelligence is a new research direction, which integrates the latest achievements of artificial intelligence and information technology, greatly emphasizes on the Internet means intelligent application of data mining technology in the online intelligence research has a very important position. This paper presents a project-based collaborative filtering algorithm hierarchical similarity. Users to take advantage of some of the projects marked tags and project categories were automatically extended, to establish a hierarchy of all projects, and then use similar items tag hierarchy established between computing projects. Experimental results show that compared with traditional collaborative filtering algorithm, the ability of collaborative filtering algorithm based on similarity of item level can significantly improve the recommendation system to handle large data presented in this paper.

Sentiment dictionaries or lexicons are core elements for “bag-of-word” approaches of opinion mining or sentiment analysis. Rather than using general-purpose sentiment dictionaries, domain-specific sentiment lexicons can contribute to improve performance because they can reflect domain specific terms and meanings. This paper presents four domain-specific sentiment dictionary construction methods for opinion mining, and describes performance evaluation results using a practical data set. The comparison subjects of this research include SO-PMI (Semantic Orientation from Pointwise Mutual Information) and three term frequency-based methods with different term polarity measures. To evaluate the performance of four different methods, a movie review data set from a representative Internet movie community site, IMDb (Internet Movie Database) is collected using a web crawling program, and is analyzed using R programs. Based on training data set, domain specific sentiment dictionaries are constructed using four different methods, and are compared their performance of sentiment analysis. The experimental results show that domain-specific sentiment dictionaries are working better than general-purpose dictionaries except one genre, „animation‟. Also, term frequency-based approaches show better performance than SO-PMI.

Data-driven decision in big data era is becoming ubiquitous in electronic grid. In particular, daily collected power consumption records enable workload aware device clustering, which is crucial for critical domain applications such as device functionality identification. In this paper, we propose a load pattern window aware method for clustering power supply devices. Our approach overcomes the drawbacks in existing works, such as fuzzy based clustering, K-means based clustering and neutral network based clustering. After investigating the large scale records from power supply devices, our approach partitions device records into disjoint time intervals with parameterized window size, which indicate the load pattern feature for a period of time given a specific device. Devices are then decomposed into a mixture of these features, and those devices with similar dominating features are grouped together. The experimental results demonstrate the effectiveness and efficiency of our solution based on the real data collected from power grid in China.

Aiming at the problem that the particle in the traditional particle swarm optimization algorithm is easy to fall into the local optimum and the convergence rate is slow, this paper proposed an improved particle swarm optimization algorithm. In particle swarm optimization algorithm, the advantages and disadvantages of the algorithm is directly decided by the performance of the particle, the paper introduced the chaos mechanism, enhance the ergodicity and particle will be quantized in the solution space, on the premise of ensuring diversity of solution, the particle get better global search ability. Meanwhile, based on the problem of slow convergence speed of the algorithm in the late, on the one hand to dynamically adjust the inertia weight of impact speed, makes the particle movement speed tend to be reasonable, on the other hand, using k-means algorithm to optimize progeny particle and get more reasonable clustering center, make the algorithm fast convergence. Experiments show that using improved Particle Swarm Optimization algorithm with high precision, strong stability and fast convergence.