Prediction Of Mobile Internet User Perception Satisfaction Based On S1-MME Interface Data Of LTE Signaling Systems

With the rapid development of mobile internet,the scale of mobile internet users is getting bigger and bigger.Users no longer concern about factors such as voice quality and charges,they have higher requirements for operators' network quality.Many users complain because of the problem of network service quality.Operators usually use market survey and users' complaint data to learn the user perception satisfaction,but these methods have not only low efficiency but also limited user range.Meanwhile,the traditional methods based on KPI(Key Performance Indicator)for evaluating user perception have strong subjectivity and poor accuracy.Therefore,it is very meaningful for operators to evaluate user perception satisfaction with network quality accurately and quickly.In this thesis,according to the S1-MME(S1-Mobility Management Entity)interface data of LTE(Long Term Evolution)signaling systems and the actual complaint data provided by mobile operators,a prediction scheme of mobile internet user perception satisfaction based on Bi-LSTM(Bidirectional Long Short-Term Memory)neural network is designed and implemented.Firstly,the S1-MME interface data is processed into feature sequences,to which the complaint or non-complaint category labels are added.And the 88,000 labeled samples are divided into a training set and a test set.Secondly,because of the processed feature sequences are time series data,in order to select the most suitable neural network to establish classification model of user perception evaluation,the application of LSTM(Long Short-Term Memory),MLP(Multi-layer Perceptron)and Elman in the classification task of the wall-following robot navigation data which is time series are researched and compared.The experimental results show that the classification accuracy of LSTM is better than that of the latter two.Then the classification model of user perception evaluation based on Bi-LSTM is implemented on Tensor Flow platform.After training,the model is evaluated by using a 5-fold cross validation method.At the same time,the classification threshold of model is adjusted,the recall is improved under the condition that precision is slightly reduced.The precision of the final prediction model reaches 0.715 and the recall is raised to 0.607.Finally,the prediction of mobile internet user perception satisfaction based on the classification model is implemented.The main content of this thesis includes four chapters: The research background and the status of research at home and abroad of the topic are introduced in chapter one.In chapter two,the requirement of the prediction of mobile internet user perception satisfaction is analyzed,moreover,the S1-MME interface data and complaint data are processed,and then samples are generated.The application of LSTM,MLP and Elman in the classification task of the wall-following robot navigation are researched and compared in chapter three.The classification model of user perception evaluation based on Bi-LSTM is implemented on TensorFlow platform,and it is used to predict the mobile internet user perception satisfaction in chapter four.