Analysis And Application For User’s QoE In Network Video Environment

Nowadays with the rapid development of Internet technology, the emergence of a large number of video diversity services leads to user’s growing demand for high video quality. In order to provide better service and improve the user’s experience of watching videos, video service providers are now working hard on various indicators to improve user’s Quality of Experience(QoE). Therefore, how to adjust and balance each index to predict QoE has become a research hotspot in the current sense. This paper carries out a series of research from the points of data analysis, modeling and prediction.The paper first introduces the data collection, data preprocessing and other preparatory work. In this period, a great deal of data related to videos are obtained. In order to analyze the influence factors of QoE, several work are carried out such as data cleaning, discretization and statistical analysis to screen out the appropriate factors. The decision tree model is selected for QoE prediction after comparing with other classical machine learning algorithms. Specifically, the main research work of this paper has the following three points:Firstly, the paper takes the information gain maximization method for discretization of continuous data in the correlation analysis with information gain. At the same time, the existing fast discretization algorithm is improved. By categorizing and integrating the data containing repeating values, the improved fast discretization algorithm can be applied to any type of data set and increase the accuracy of the discretization. Meanwhile, it is proved that the improved fast algorithm can accelerate the speed of structuring the tree and achieve the same accuracy of the original algorithm by applying the method to the two-part discretization.Secondly, in order to simplify decision tree C4.5 model, the stop condition is improved. Based on the trade-off between model complexity and accuracy, the minimum amount of data subsets to be split is determined to achieve the simplest model under the condition with the guarantee of the accuracy. Experiments show that the complexity of the decision tree and the consuming time can be reduced greatly, and in this way the model can be applied to the data with much larger amount.Finally, several improvements are proposed according to the criterion of feature selection in the decision tree’s modeling process. By introducing the feature dispersion index combined with the original information gain rate index, the new selection criteria is established through considering the dispersion and distribution of the data itself. It is proved that the new standard can make the feature selection more accurate and improve the prediction accuracy of the decision tree model.At the end, the paper tests the accuracy, complexity and consuming time of the model compared with other models, and the results verify the performance of the design scheme and prove that the improved algorithm can increase the prediction accuracy of QoE.