网络视频环境中面向用户QoE的分析与应用

发布时间：2018-01-01 13:35

本文关键词：网络视频环境中面向用户QoE的分析与应用　出处：《南京邮电大学》2016年硕士论文　论文类型：学位论文

【摘要】：互联网技术飞速发展的当下,视频多样性服务也大量涌现,用户对于视频质量的需求也随之增长。为了更好地服务用户,提升用户体验,服务商致力于研究各方面指标以改善用户体验质量(QoE)。因此,如何通过调整、权衡各指标来预测QoE成为了当下的研究热点。本文从数据分析、建模预测的角度出发,展开了一系列研究。本文首先进行了数据搜集、数据预处理等前期准备工作,从而获得了大量与视频相关的指标数据。为了对可能影响QoE的因素进行分析,本文进行了数据清洗、离散化、统计分析等工作,从而筛选出合适的影响因素。通过对比多种经典机器学习算法,选取决策树模型对QoE进行建模预测。具体而言,本文的主要研究工作有如下三点:首先,在利用信息增益进行相关性分析时,采用信息增益最大化方法对连续型数据进行离散化处理,并对现有快速离散算法进行改进。通过对含有重复值的特征数据进行重新整合归类,使快速离散算法能适用于任何类型的数据集,提高了离散化的准确率。与此同时将改进快速算法运用到构建树时的二分离散中,加快了建树速度,又达到了与原树相同的预测效果。其次,为了精简C4.5决策树模型,对停止条件进行了改进。通过将模型复杂度与准确率权衡比较,共同决定待分裂子集数据量的最小值,使决策树在保证一定准确率的前提下达到最精简模式。通过实验验证这样改进会使得模型的复杂度、算法耗时大大降低,从而能够适应更庞大的数据量。最后,针对决策树建模过程中的特征选取准则进行了改进。通过引入特征离散度指标,与原始信息增益率指标相结合,综合考虑数据本身的离散程度和分布情况,构建新的选取标准。通过实验证明新的选取标准使得特征选取更为准确,同时可以提高决策树模型的预测准确率。本文最后通过与其他模型进行对比,对改进模型准确率、复杂度、耗时等多方面进行测试,验证了所设计方案的性能以及改进算法在QoE预测准确性方面的提升效果。
[Abstract]:The rapid development of Internet technology at present, the diversity of services is the emergence of a large number of video users, along with growth for video quality requirements. In order to better serve customers, enhance the user experience, the service provider is devoted to the study of the indicators to improve the quality of user experience (QoE). Therefore, how to adjust, to predict the QoE become a research the focus of the present balance each index. This paper from the data analysis, modeling and forecasting perspective, launched a series of research. This paper carried out the data collection, data processing and other preparatory work, to obtain a great deal of data associated with video. In order to analyze the influencing factors of QoE, this paper has carried on the data cleaning, discretization, statistical analysis and so on, so as to select the appropriate factors. By comparing various classic machine learning algorithms, then the decision tree model is used to model QoE Forecast. Specifically, the main research work of this paper has the following three points: first, the correlation analysis based on information gain, the information gain maximization method for discretization of continuous data, and the existing fast discrete algorithm. Based on the characteristics of data containing duplicate values were re classified, the fast discrete algorithm can be applied to any type of data sets, to improve the accuracy of the discretization. At the same time the improved fast algorithm using two discrete to construct the tree, accelerate the speed of tree and predicted the same effect as the original tree. Secondly, in order to streamline the C4.5 decision tree model, to stop the condition was improved. The model complexity and accuracy tradeoffs, jointly determine the minimum amount of data to be split subset, the decision tree to the issued under the premise of ensuring accuracy Reduced complexity model. Experiments show this improvement makes the model, computation time is greatly reduced, which can adapt to a large amount of data. Finally, according to the criterion of feature selection decision tree modeling process was improved. By introducing the characteristics of dispersion index, combined with the original information gain rate index, considering the discrete the extent and distribution of the data itself, the criteria for the selection of new building. Experiments show that the feature selection criterion is more accurate, and can improve the prediction accuracy of the decision tree model. Finally, by comparing with other models, the improved model accuracy, complexity, time and other aspects of the test. To verify the design scheme of the improved algorithm performance and forecasting accuracy improvement in QoE.

【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP393.09

【相似文献】