Huge amounts of telecom data are generated every day, so how to extract useful information from the data is one of the data mining problems. Because different applications need different clusters, sometimes a single K-means cluster algorithm cannot generate userspecified K-clusters. An adjustable multitimes clustering mining method is proposed. A big value K was used in the K-means clustering algorithm for the first time, and K clusters were obtained. They were used to select the number of the clusters and the initial centers of the clusters for the second time. The experimental results show that our method is effective, and it can be applied to mining different amounts of clusters and big data analysis.

Big data will have a profound impact on economics, society and life in the near future, and the research on integration and storage of big data may play an important theoretical and practical role in promoting and deepening the application scope of big data. It utilized the data storage structure of the distributed file system named as HBase and the basicelement of Extenics to integrate the heterogeneous data sets, and then stored the processed data set in HBase database. The new data set, which was obtained by extracting the typical characteristics as well as their value of data, especially the semistructured and unstructured data, provides not only a new way for the analysis and interpretation of data but also the research ideas and strategy generation for the professional issues from the perspective of big data.

A multiobjective optimization algorithm inspired from cloud model and using favour ranking is introduced. The innovation of the algorithm lies in the estimation of good solution regions and new solution production according to the cloud model theory. The algorithm used information obtained during optimization to build the cloud model for good solution regions, and estimated three digital characteristics of the cloud model by backward cloud generators. Afterwards, forward cloud generators were used to generate current offsprings population according to three digital characteristics. The population with the current population and current offsprings population was sorted using favour ranking, and the best individuals were selected to form the next population. Regarding a set of benchmark functions, the proposed algorithm was tested and compared with some other algorithms. The experimental results show that the algorithm is effective  in the benchmark functions.

To help enterprises to obtain timely, accurate and reliable intelligence information under the environment of big data, it proposed how to build an enterprise competitive intelligence system, based on big data. It used web crawlers of focused crawler works to collect intelligence information, and employed the Hadoop-based KNN algorithm to process and classify the intelligence information via the B/S (Browser/Server) framework, which can solve the problem of high time complexity of the KNN classification algorithm in a big data environment and other issues. The system also allows users some customization, making it more individualized.

The aim of clustering is to help people find and recognize the unknown world, so as to accumulate knowledge for us in real life. Clustering analysis is an important part for the majority of researchers in unsupervised leaning, and is usually used as an analysis tool to explore the unknown data and its regularity for many cross subjects. It analyzed the procedure of clustering, and briefly surveyed the related achievements. Moreover, the problems of clustering algorithms in processing various data types, high dimensional data, unbalanced data were concluded, and the expansibility and the selection of evaluation index for algorithms were also discussed in detail. At last, some directions for future research were proposed. The above work can give valuable reference to further studies of clustering and data mining.

By combining the PageRank algorithm with the features of intensive noisedata to improve the noise-data recognition rate of DBSCAN in environments with intensive Noise-Point distribution, it structured the inner-cluster mapping function for voting, and proposed the inter-cluster voting noise recognition algorithmNoiseRank. Experimental results show that in environments with intensive NoisePoint distribution, the Noise-data recognition rate of NoiseRank is much higher than that of DBSCAN.

Collaborative filtering has been widely used in the recommendation system, but the traditional  algorithm has some limitations, such as inability to adapt to the sparsity of user-item rating matrix data sets well, failure to consider the classification of item, users-scores, interest change over time and other factors when calculating the similarity of the items. Regarding these limitations, it proposed a big data set hybrid dynamic collaborative filtering algorithm, based on the traditional collaborative filtering algorithm. When calculating the similarity of items, time decay functions were introduced in the algorithm, which considered both the similarity of items, scores and items classified. The weights of project integrated similarity could be adjusted automatically. In the algorithm, some improvements have also been made in similarity computing and the selection of the neighboring items. To verify the effectiveness of the algorithm, experiments were done on movielens data sets. Experimental results show that the algorithm is better than the traditional recommendation algorithms.

It proposed a novel secure cloud storage model (ASOM), based on HDFS. Through isolating the metadata from the physical storage and the communication between Data nodes and metadata subservers, the ASOM model can guarantee secure metadata management. The main advantages of our approach include avoiding the superposition of complex security policies and the mistrust between the users and the platform. Furthermore, our security storage service can be easily integrated into the cloud computing environment.

Cloud computing has the nature of  being dynamic, virtual and open since it was used, and all kinds of largescale cloud security incidents make the safety of cloud environments frequently questioned. To ensure the security of cloud environments, it proposed an intrusion prevention framework model, based on trusted computing in cloud environments, by combining intrusion prevention technologies and trusted computing ideas. The model began with the principle of intrusion prevention with access to behavioral characteristics. Then, these features were gradually normalized, and the weight of each feature was determined to obtain user nodes' credibility. Next, it used a variety of cloud cluster server engines to detect defense and make integrated decision analysis and cluster analysis, enabling the cloud to make timely fast intrusion prevention, which avoids the drawbacks of the traditional intrusion prevention, such as minding only their own business, lagging behind in detecting and preventing attacks. The model provides cloud users with the maximum intrusion prevention services, and ensures that the cloud can withstand attacks, making the cloud and cloud users secure.

In view of the existing defects of RFID authentication protocol, based on the Hash function, and the low efficiency of the security protocol authentication in the application of the Internet of Things,  it proposed an improved RFID security mutual authentication protocol, based on the Hash function. This protocol can protect data privacy, and prevent replay attack, privacy track, and spoofing attack. Compared with other protocols of this kind in security and performance, this protocol, which uses Label certification mark Tuse, Tstore and dynamic secret value S, can prevent desynchronization attack, has higher efficiency, and is suitable for low-cost RFID systems.

Aiming at the high requirements for security of cloud computing services, it  proposes an anonymous authentication protocol in cloud computing, and designs a model that suits identity authentication. The temporary identity of the user was constructed by using bilinear pairings. Besides, the nonce was used to replace time stamps to avoid the problem of clock synchronization. The protocol implements mutual authentication, based on the calculation difficulty of  the Discrete Logarithm Problem and the irreversibility of hash function. The protocol is efficient and has the characteristic of high security. It can be applied in distributed cloud environments, which need to protect the users privacy.

Cloud computing and super computing are the products of information age. Aiming at the users needs in their use of CAE software and its complex business for high performance computing, it proposes the architecture design and the key technology of implementation of the CAE software system, based on integration of  cloud computing and super computing, and gives an example to illustrate that.

It studies the task scheduling in cloud computing, and proposes a hybrid scheduling algorithm(ACOSA) combined with ant colony algorithm and simulated annealing algorithm for the MapReduce programming framework of cloud computing. This algorithm aims at minimizing the scheduling time and introduces the task and resource matching factors and load balance. Firstly, the ant colony algorithm was used to get the optimal solution to a set of tasks and resources. Then, the path was optimized, and the pheromone of solution was updated by  the simulated annealing algorithm. Lastly, they were recompiled by extending Cloudsim cloud computing simulation platform, and the ACOSA algorithm was achieved. The experimental results show that the algorithm has a good performance in scheduling time and load balancing.

Regarding energy consumption for different degrees of nodes in a wireless sensor network, it proposed the energy cluster head selection mechanism, based on the cost of energy. In this mechanism, the first common node was seen as the cluster head node. Nodes, necessary node energy consumption, and the current energy surplus value were computed. With the proposed concepts and formulas, the above energyrelated value was converted into the value of energy loss speed of the selected clusters at measurable sensor nodes. Simulation results show that the algorithm has a better cluster head selection, and it can effectively extend the network life cycle.

Concerning the problem with the recognition and normalization of time expressions in texts, aiming at the diversity and unstructured forms of time expressions, it proposed the idea of describing temporal elements to divide the types of time expressions and their forms of normalization.  With the method that combined regular expressions with Trie tree structure, it built the recognition tree of time expressions, which could recognize time expressions automatically. Finally,  it proposed the normalization algorithm and correction algorithm to deal with the recognized results. The results are pretty good.

The key problem to solve in a sentiment analysis of texts is the sentiment polarity classification. Based on the analysis of various factors affecting sentiment classification of texts, it built the sentiment lexicon, extracted affective characteristics, and weighted sentimental features. Then, it used support vector machine (SVM)  classifier for emotion recognition and text classification. Finally, it performed the classification model with the corpus data sets on the single platform and the Spark distributed computing platform to analyze its classification accuracy and time cost. The experimental results verify the effectiveness of the text sentimental polarity categorization model on the single platform and on the spark distributed computing platform.

To meet the needs for storing and quick retrieving mass unstructured temporal information, it proposed using the distributed and unstructured database HBase, which was on the Hadoop platform, to store temporal data. Then, it built the temporal data storage model with the store unit as the temporal set, and designed a Multi level indexed Distributed Hash Table (tDHT) algorithm to realize the retrieval for the temporal attribute value of temporal column quickly and efficiently. By mapping from temporal attribute value to the twodimensional space, the conversion from temporal data to space object was achieved, the temporal data area was divided by using the processing method for spatial data, Multi level temporal data subareas were generated, and the Multi level indexed DHT directory was constructed, which was stored by HBase, using the methodology of DHT. The experiment results show that the index strategy can achieve a good performance, and it can be used to accelerate temporal data retrieval in the HBase table  to a certain extent.

The uncertainty and asynchronous nature in the implementation of multithreading makes it fairly difficult to test the correctness of the concurrent program. To improve the efficiency, it proposed a method to test the correctness of concurrent programs, based on the nondeterministic test method. Through intensifying concurrent programs to compete for resources, the potential concurrency errors were found and the correctness of concurrent programs was tested. The experimental results show that with this method, the efficiency of testing concurrent correctness is validly improved. And that the errors in the concurrency program can be found with more efficiency.

It discussed the core ideas and programming model of Storm, and analyzed its working mode and application method. Finally, it implemented the performance and horizontal scaling test of a data processing system based on Storm. The experimental results show that the performance and scalability of Storm is superior to that of the traditional data processing system.

K nearest neighbor query algorithm is one of the commonly used algorithms in massive spatial data query. First, it construct an index of largescale spatial data by Kd-Tree, and hierarchical division of the search space. Then, it used the k nearest neighbor query to ensure the efficiency of the search. However, the traditional Kd-Tree construction has two drawbacks: the use of test data points are required for each k nearest neighbor query back to the root, thus affecting the efficiency of the query; Kd-Tree uses the splitlevel domain for the space division of space into cubes (twodimensional data are rectangular), extra space appears in polygonal space at the intersection of judgment, making the comparison of data unnecessary,   thus affecting the efficiency of the query. Regarding these two shortcomings, it proposed the corresponding improved algorithmRB algorithm. Experimental results show that the algorithm has a higher query efficiency than the traditional KD algorithm.There are two main contribution from this paper: (1) This paper construct a quickly create Kd-Tree indexes to support queries KNN akgorithm to classify largescale data. (2) RB algorithm is proposed to improve the traditional Kd-Tree index construction method,and improving query efficiency for KNN algorithm.

To replace traditional paper shelf labels, it proposes a power efficient bimodal electronic shelf label system, based on the technologies of the Internet of Things and wireless sensor networks. Focusing on the power efficiency, the proposed electronic shelf label adopts an electrophoretic electronic paper display and a Bluetooth module as the display and communication modules respectively. Moreover, this label has two working states: wake-up state and sleep state, enabling it to switch the working state under the server commands, to report label status, and to update display contents with low power consumption. The experimental results show that this system can meet the basic requirements of application scenarios. It has an advantage of comparatively low running costs over traditional paper labels.

It proposes a sleep method of the wireless electronic shelf label system using relative time, based on the Internet of Things. Then, it provides a device model for electronic shelf labels controlled by relative time, the system workflow, a timesharing queuing algorithm, and a relativetime sleep algorithm. The experimental results show that the effective use of relative time helps to manage a large number of asynchronous electronic shelf labels working together. The working status of each label is accurately and simply controlled via a server that commands the time of sleep, wakeup, and communication of each label, thus increasing each label's power efficiency.

Aiming at more efficient and reliable storage and management of RFID tag data, it designed a management system of RFID tag data, based on cloud storage, and proposed a data integrity check algorithm, based on Hash function. According to the field of the RFID tag data, the algorithm generated a unique checksum to confirm whether the data stored in the RFID tag was damaged. If RFID tag data had been damaged, the corresponding RFID tag data can be read via any network from the cloud storage system to recover its data.