Conversational Flow Classification Based On Multimedia Service QoE Model

With the rapid development of Internet technology,multimedia technologies progress faster and have a wide range of types.Live video,2D games,3D games and other high-speed media,high-throughput,multi-user simultaneous services generate in the network a large number of mixed flows.From the perspective of network service providers,how to identify and classify the mixed flows efficiently with limited network resources,andto provide the user’s QoE(Quality of Experience)is very important.The quality of service(QoS)and Quality of Experience are used to characterize the relevant parameters and user’s perception of the network.The multimedia traffic can be divided into four classes according to the service type: interactive class,background class,streamingclass,conversational class.Different types of QoS parameters have different degrees of impact.This thesis focuses on conversational class.By considering its low latency and jitter,high throughout,we analyze the QoS-QoE model of mixed session flow and the relationship between the probability distribution of Mean Opining Score(MOS)value and QoE under different QoS features.The most obvious feature probability distribution in the form of network flow characteristics is put into the original feature set of the conversational flow,leading to the classification of four types of conversational flows: QQ video,2D game,3D game and Skype voice.Thisthesis combines conversational flow service features,selects three QoS representative parameters: delay jitter,throughput and average packet interval,and analyzes the existing models.In this thesis,the QoS model containsthree parameters,and the weight andinfluence factor of the parameter are also set.The relationship between the QoS characteristic and the probability distribution of the MOS value is obtained.Then,the thesis does the statistical analysis,andadds the MOS probability distribution into the classification feature set to identify the mixed applicationsof the conversational class.The experimental results show that the improved QoE model can enhancethe accuracy of multimedia mixed traffic classification.At the same time,we can get the specific QoE model function expression,thus verifying the correctness of the idea.