Research On Multimedia QoE Prediction Based On Broad Learning System

In recent years,with the rapid development of Internet technology and the rapid popularization of terminal devices(such as mobile phones,tablets,PCs and high-definition Internet TVs),the demand for multimedia services which have become indispensable in people's entertainment life has increased.With further development,the quality of services and user experience have gradually become the focus of attention of users and various service providers.Quality of Experience(QoE)is an important indicator for evaluating multimedia services.However,QoE is a complex evaluation indicator that is affected by many factors,not only by objective network parameters,but also by user subjective factors.Therefore,accurately predicting the QoE of multimedia services,and then improving the problem in a targeted manner,improving user experience,and retaining different user groups are the key concerns of multimedia service providers.In view of this,this paper mainly studies the multimedia QoE prediction model based on Broad Learning System(BLS).The main research work and innovations of this paper are as follows:First,perform data preprocessing on the IPTV data set.Due to many uncontrollable factors in the data collection process,the data set contains some abnormal data.These abnormal data will affect the subsequent modeling process,so these data must be deleted or filled;then standardize each feature in the data to remove the restriction of the data unit and facilitate weighting or comparison between each feature.Subsequently,the feature engineering of data is completed,including two aspects: the selection of objective features and the extraction of subjective features.The selection of objective features was based on correlation coefficients and information gains.Not only did it consider the linear relationship between data from multiple aspects,but also taking into account the non-linear relationship.The extraction of subjective features is based on the more common pause behavior in the user's viewing,and the relationship between features and QoE is considered from the perspective of user experience.On the basis of the above data preprocessing and feature engineering,the network structure of BLS is built using Python,and then QoE predictive modeling is performed using BLS.The detailed algorithm flow is given.The experimental results show that the BLS-based prediction method has better prediction performance than other prediction methods,and is both effective and efficient.Then,Uncertainty Autoencoder(UAE)is used for unsupervised representation learning,which serves as the pre-feature extractor of BLS.UAE makes up for the shortcomings of insufficient feature learning in BLS.If the performance of BLS does not achieve the expected results,more network nodes can be added.Using the incremental learning algorithm of BLS can quickly obtain new network weights.The incremental learning algorithm improves the flexibility of the overall model.Based on the advantages and disadvantages of the two,UAE and BLS are combined as a joint model for QoE predictive modeling,and the advantages of the joint model are explained theoretically from the perspective of learning objectives,workflow and model structure.Finally,the UAE model is built using the Py Torch learning framework,and the module design and experimental process in the implementation process are introduced in detail.On the basis of specific experimental parameters,the performance of UAE-BLS was verified on different data sets,and the advantages and applicable scenarios of the model were clarified.