【作者】 彭玉静；

【导师】 赵晶玲；

【作者基本信息】 北京邮电大学 ， 计算机技术（专业学位）， 2013， 硕士

【摘要】 移动互联网技术的日趋成熟,应用种类的不断丰富,用户规模的快速增长,加速了数据量的增长。在这些海量的数据中,隐藏着大量与业务性能相关的信息,通过这些信息能够分析出各种业务性能之间的差异情况。在这个分析的过程中,数据挖掘技术扮演了非常重要的角色。利用数据挖掘中的相关技术,能够从海量数据中挖掘出有效的性能特征信息,并对具有不同性能特征的业务进行归类分析,以实现对移动互联网业务性能的分析,从而有利于平衡网络的负荷和提高网络资源的使用率。本文设计了一个用于研究移动互联网业务性能的分析系统。该分析系统包括三个模块：数据信息采集模块,数据预处理模块和数据挖掘模块。数据信息采集模块利用JPCAP中间件技术和数据库存储技术,对业务中包含的有效信息数据进行识别、选择、采集和存储等操作。数据预处理模块包含对存储的业务数据的清洗、集成和变换等过程。数据挖掘模块,是利用数据挖掘中的k-means聚类分析算法对业务的网络性能属性进行聚类分析操作。传统的k-means聚类分析算法是采用随机的方法确定初始质心,这种确定方法会直接导致聚类的结果存在误差且带有随机性。因此本文对算法中初始质心的确定方法进行了改进,实验分析验证该改进后的算法是可行的并且聚类分析的结果是有效的。最后,本文将该改进后的算法应用到了分析系统中的数据挖掘模块中。更多还原

【Abstract】 The mobile Internet technology increasingly mature, the rapid growth of application types and constantly enrich the user scale, and accelerate the growth of the amount of data. A large number of hidden information associated with business performance in vast amounts of data, which can be by analysis of the difference in a variety of business performance. Data mining technology plays a very important role in the process of analysis. Using data mining technology, extract the information with effective performance characteristics from massive data, classify and analyze the business with different performance characteristics at the same time, achieving the performance analysis of the mobile internet business, which is conducive to the balanced network load and improve network resource utilization.In this paper, a system used to study the performance of the mobile Internet business analysis is designed. The analysis system consists of three modules:data acquisition module, data preprocessing module and data mining modules. Handle the effective information with identification, selection, collection and storage in data information acquisition module using JPCAP middleware technology and database storage technology to handle. The data preprocessing module contains processes on the effective information data including purification, integration and transformation. Data mining module, using the k-means clustering analysis in data mining algorithm to cluster analysis of operating performance attributes of the business. The traditional k-means clustering analysis algorithm uses the randomized method to determine the initial centers, so it will cause the results of clustering error and randomness. Therefore the method of determining the initial centers is improved, and verified by experiments that the improved algorithm is feasible and the results of cluster analysis are valid. Finally, the improved algorithm is applied to data mining analysis system module.更多还原