Research On The Prediction Of User Satisfaction Based On Neural Network

With the rapid development of mobile Internet technology,the Internet applications based on mobile communication were increasing gradually,which brought many kinds of conveniences to people's life.In recent years,the consumer service platforms based on mobile communication had gradually entered mass life.Taking the yelp of the largest review website in the United States as an example,by the end of 2018,the total number of user review informations collected by the yelp website had exceeded 177 million,the average number of active users per month had been about 7 million,and the user size of the platform had continued to grow.In order to occupy the advantage in such a huge application market competition environment,the review applications needed to innovate their own product technologies from the angle of improving user satisfaction,in order to enhance the user's software application experience.Based on the data sets of yelp,this paper proposed a user satisfaction prediction method based on user preference mining.The main research processes and conclusions included the following three parts:(1)Based on the user consumption data set and shop information data set published on the yelp website,a series of data preprocessing work had been carried out,including raw data segmentation,data cleaning,data normalization,and text data preprocessing.Considering that the shop service has regional characteristics,the experimental data were selected through the way of geospatial data screening.(2)Based on the analysis of consumer behavior,this paper proposed two modeling methods to mine user preferences.The first method was based on shop type labels and user consumption frequency,the EM(Expectation-Maximization)clustering method of GMM(Gaussian Mixture Model)was proposed to generate user consumption behavior vectors,which then quantified the consumer behavior patterns and provides user behavior characteristics for the following research work.The second method was the RNN(Recurrent Neural Network)user preference mining model based on attention mechanism,and based on the RNN,the improved LSTM structure was used for user preference prediction.After experimental verification,it was shown that the attention mechanism based user preference mining model had better prediction effect than the original RNN model,and the accuracy was 0.746.(3)This paper improved the input layer of artificial neural network based on the user preference characteristics which were defined by the output result of the user preference mining model.The spatial aggregation degree of shop location was obtained by using DBSCAN spatial clustering method,and the shop level feature and user level feature are then used as the input layer of neural network model to predict user satisfaction.Finally,the improved neural network user satisfaction prediction model was compared with other traditional models,and the results showed that the proposed user satisfaction prediction model was better than the traditional model,and the accuracy was 0.814,and it was verified that the prediction results of user satisfaction were closely related to the user's preference characteristics.