基于t分布混合模型的网络多媒体业务流分类方法研究

发布时间：2018-06-21 05:29

本文选题：流分类 + 多媒体业务　；参考：《南京邮电大学》2017年硕士论文

【摘要】：随着互联网近年来的不断普及和发展,新型的网络业务层出不穷。然而,网络业务的增长给网络流的分类和管理工作带来了极大的困难。多媒体业务(如音频业务、视频业务、网页浏览等)作为互联网业务中最重要的组成部分之一,占据着最大一部分的网络流量。因此,对多媒体业务的流分类是一个重要的研究内容,有助于网络业务提供商(ISP,Internet Service Provider)为用户分配合适的网络资源,保障业务的服务质量。本文主要针对6种多媒体业务进行了细粒度的分析:在线音频类业务(即时语音通信和网络音乐)、在线视频类业务(直播和非直播)和网页浏览(搜索引擎和网络新闻业务)。本文对上述多媒体业务进行QoS(Quality of Service)特征的研究,并对这些业务的流样本进行了建模工作。具体研究工作如下:针对传统高斯混合模型(GMM,Gaussian mixture model)容易受到数据样本边缘值和离群样本的影响,本文用t分布混合模型(TMM,student's t-distribution mixture model)替代原有的高斯混合模型,并使用EM(Expectation Maximization)算法对6种多媒体业务进行了建模工作。之后,提出了有限t分布混合模型(LTMM,Limited student's t-distribution mixture model)。该模型可以降低EM算法的迭代次数,有效减少建模时间。此外,本文从理论和实验的角度将提出的混合模型与传统的K-Means和高斯混合模型进行了复杂度和分类准确度的对比,从而验证了两种模型的合理性。本文结合两种提出的混合模型,使用改进的半监督流分类方法对多媒体业务进行了分类工作。该方法可以分为数据预处理、数据建模以及数据分类三个部分。本文对该方法的聚类中心、细分类算法等内容进行了深入地探讨。实验结果表明,使用本文的半监督流分类方法对于多媒体业务可以获得较高的分类准确度。
[Abstract]:With the continuous popularization and development of the Internet in recent years, new network services emerge in endlessly. However, the growth of network traffic brings great difficulties to the classification and management of network flows. As one of the most important components of Internet services, multimedia services (such as audio services, video services, web browsing, etc.) occupy the largest portion of network traffic. Therefore, stream classification of multimedia services is an important research content, which is helpful for ISPU Internet Service provider to allocate appropriate network resources to users and ensure service quality. In this paper, six kinds of multimedia services are analyzed in detail: online audio services (instant voice communication and network music services), online video services (live and non-live) and web browsing (search engine and network news services). In this paper, the QoS quality of service features of these multimedia services are studied, and the flow samples of these services are modeled. The research work is as follows: in view of the influence of the edge values of data samples and outlier samples on the traditional Gao Si mixed model, the t-distribution mixed model (TMM) is used to replace the original Gao Si mixed model. The EMN expectation maximization algorithm is used to model six multimedia services. After that, a finite t distribution mixed model is proposed. This model can reduce the number of iterations of EM algorithm and effectively reduce the modeling time. In addition, the proposed hybrid model is compared with the traditional K-Means and Gao Si hybrid models in terms of complexity and classification accuracy from the theoretical and experimental points of view, thus validating the rationality of the two models. In this paper, an improved semi-supervised stream classification method is used to classify multimedia services. The method can be divided into three parts: data preprocessing, data modeling and data classification. In this paper, the clustering center and fine classification algorithm of this method are discussed. Experimental results show that the proposed semi-supervised stream classification method can achieve high classification accuracy for multimedia services.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0

【相似文献】