Research On Key Technology Of IPTV User Experience Improvement Based On User Behavior Protrait

With the development of multimedia technology and the integration of the three networks,the Internet Protocal Television(IPTV)is gaining more and more attention and favor.IPTV integrates multiple technologies such as Internet technology,communication technolog,and internet technology.It provides diversified and user-friendly interactive services based on existing IP networks.With the increase of IPTV users,how to improve user satisfaction and more accurately predict the quality of user experience has become the core competitive goal of retaining users and improving user stickiness.Based on the above challenge,this article has carried out three aspects of research work.The specific work and contributions are as follows:First,we propose a feature selection method combining subjective and objective features.Features are processed from aspects of both selection of objective features and construction of subjective features.The selection of objective features is mainly to calculate the correlation between the feature field and the prediction target from the two aspects of correlation analysis and information gain.The object is to fully retaining valid information.The construction of subjective features starts from user preferences.It is committed to reveals the relationship between the quality of the user's viewing experience and the user's personal preferences and program popularity.The feature selection method proposed in this paper combines subjective and objective features,which can reduce the complexity of the model while reducing the complexity of the model.At the same time,it can take into account the effects of time preference and program preference,and help subsequent models to better train and predict.Secondly,we research the personas method based on dynamic time warping(DTW)and collaborative filtering.We analyze the user's viewing behavior from two levels to finally complete the user portrait.On the one hand,we analyze the perspective of user viewing time.Based on the DTW method,the distance between the user's viewing intensity and the typical viewing intensity is calculated to complete a portrait.On the other hand,we analyze the perspective of user viewing content.Based on the collaborative filtering method,the value of the Jaccard coefficient between user preferences and group preferences is calculated.The result of the portrait is modified once to complete the final portrait.Finally,the user group is divided into young users,elderly users and atypical users.This method can effectively grasp the characteristics of the group to provide advantages for subsequent model prediction.Finally,we propose a prediction model of IPTV user experience quality based on improved neural network.Based on the front-to-back dependence of user behavior,we choose the long-short term memory network(LSTM)model to better grasp the dependencies between data.Aiming at the limitation of LSTM itself in computational complexity,the LSTM-SVM model and the LSTM-DT model were proposed respectively.The improved model can take into account the classification advantages of support vector machine(SVM)and decision tree(DT)on the premise that LSTM takes long-term and short-term dependencies into consideration.The experimental result shows that feature engineering can effectively improve the accuracy of experimental prediction without changing the model complexity.At the same time,according to the results of user portraits,it is possible to select the prediction model based on the characteristic of targeted user.