Research And Implementation Of The Evaluation Model For Internet Video's Quality Of Experience

In recent years,Internet video industry is developing rapidly in China,and Internet video service has become one of the most important businesses of domestic Internet industry.With the continuous expansion of the number of Internet video users,the audience look forward more and more eagerly to the good video quality as well as rich video resources.Nowadays,there are many websites in China providing video service,and well-known video sites are not in the minority.The competition in Internet video industry has become more and more intense.When more than one video websites own the video resource a user loves,it is more likely for this user to choose the website providing better video quality.Therefore,more and more video service providers start to pay attention to the research of Internet video's quality of experience.However,Internet video's new media business today is totally different from traditional video service in the aspects of system architecture,user access patterns and performance requirements.Video service providers are always short of deep understanding of user behavior under new situation and don't have reasonable method to evaluate the service quality either.As a result,they have to face the fact that sometimes they spend a great deal of manpower and money to carry out the optimization of the video system and platform but receive tiny effects.Over the past few years,there have been several studies on quality of experience(QoE)for Internet video in foreign.Nevertheless,considering the differences between China and other countries and areas on cultural background,the environment of video industry and the patterns of user behavior,overseas studies can just be taken as a guide.It is still quite necessary to establish a set of methods for the evaluation of QoE for domestic Internet video.In this paper,we establish an evaluation model for Internet video's QoE with the algorithm of fuzzy decision tree based on real user access data from CNTV,a large-scale video website in China.Firstly,we preprocess the original semi-structured log data to make it structured.We also infer the information about user's location and access device according to the IP address.At last,we get a set of structured video session records which contain the information on video content,the user,and the session.Next,we use some mathematical methods to analyze this set to find some regularities,e.g.the distribution of individual video quality metric is relatively concentrated and the correlation coefficient of individual video quality metric and user engagement is small.These findings offer guidance for the modeling.Finally,we cast the evaluation model for Internet video's QoE as a classification model.Considering the characteristics of our data set,we establish our model using the algorithm of fuzzy decision tree.We further compare the prediction accuracy of our model with the model developed using C4.5 on several different datasets to test and verify the effectiveness of our model.We also prove the influence of some confounding factors,such as the content of the video,on our model's performance by comparing the prediction accuracy of our model on the universal data set and its several subsets.