Analysis Of Data And Reviews Of Mobile Apps

With the development of information technology,the functions of mobile smart phones are becoming more and more abundant.Users can install various applications(hereinafter referred to as App)according to their own needs,and can achieve real-time contact with the outside world through a convenient network.A large number of apps emerge every day in major application markets.These apps,while enriching the choice of mobile phone users,also increase the difficulty for users to choose apps.Therefore,analyzing the attributes of the App and its review text can not only provide suggestions for user selection,but also guide the developer of the App to update and develop a higher-quality App according to the user's needs.s concern.Different from the traditional sentiment analysis research,this article builds an evaluation system based on the App's numerical attributes and user comment text to comprehensively analyze the advantages and disadvantages of the App.(1)Perform descriptive statistical analysis on the basic properties of the App,and get some basic characteristics of the popular App,such as small memory consumption and low price(mainly free).(2)Conduct sentiment analysis on App's review content,try to realize multiple sentiment analysis methods under different word granularity,the selected sentiment classification methods are:(i)sentiment analysis using sentiment dictionary and custom rules;)Use the simple Bayesian and support vector machines in the traditional machine learning model to establish the sentiment classification model of the review text;(iii)Choose three deep neural network models:Stacked Bidirectional Long Short Term Memory,Referred to as Stacked Bi-LSTM);Stacked Bi-LSTM with attention mechanism;and Recurrent Convolutional Neural Network(RCNN)to analyze the review text.The results show that the rule-based sentiment classification model is limited by the sentiment dictionary and judgment rules,and its accuracy is somewhat random.Secondly,the training method using traditional machine learning does not rely on manually labeled corpus,and the text processing is more objective and reliable,and the classification effect is improved to a certain extent.Finally,the deep learning-based network model obtains the features of the text through automatic learning,realizes an end-to-end joint modeling architecture,has better text classification capabilities,and the RCNN network model that features word granularity performs best.Suitable for sentiment classification of mobile app review text,and retraining with improved data set,the accuracy rate reaches 96.74%.(3)Visually display the classified comments,analyze the advantages and disadvantages of the App,and quickly understand the user's attitudes to the App.