Research On The Recognition Of False App Reviews Based On User Behavior

With the advent of the mobile Internet era,the App software industry has begun to show a "spurt" development,and the market competition has become increasingly fierce.As a reference for App quality evaluation,the comment section has been flooded with various false comments,so it is urgent to identify App false reviews effectively to protect the rights of consumers and openers,and maintain a reasonably healthy and orderly market.This paper makes use of the user data of a certain domestic app in the real scene,starting from the two dimensions of statistical model and algorithm model,and establishes a variety of classification models and comprehensively evaluates them,aiming to provide an efficient App false comment recognition method.The main work of this paper is summarized as follows:1.Preprocessing of App user data.First,the App user data is desensitized,and the obtained data set is subjected to cleaning operations such as missing value processing,outlier processing,and data conversion to form an effective high quality data set for modeling analysis.2.Descriptive statistical analysis of App user data.Using the density map,box plot,scatter plot and contingency table test to explore the characteristics of App user data,provide guidance for subsequent analysis of influencing factors and model establishment.3.Construct a logistic regression model for App false comment recognition.Firstly,the data set is tested by multicollinearity.Secondly,the ridge regression and lasso methods are used to repair the multicollinearity problem.The L1 regularization and L2 regularized logistic regression models are established respectively.Finally,the model coefficients obtained by lasso method were used to analyze the influencing factors of false comment identification of App.4.Construct an algorithm model for App false comment recognition.Using the support vector machine,random forest,gradient lifting tree,BP neural network and other algorithm models to establish a variety of App false comment recognition models,and use the grid search method to optimize the parameters of each model,while using the model combination to integrate multiple identification models that have been established,in order to complement each other and improve the overall performance.Finally,a comprehensive evaluation index system is established to analyze the advantages and disadvantages of each model,and the gradient lifting tree model has the best recognition performance.This paper starts from the user behavior level,applies a variety of classification models to the study of App false comment recognition,and establishes relevant evaluation indicators to compare and optimize the model,aiming to provide App developers with a way to effectively identify false comments,maintain the market economic order and the interests of consumers has certain practical significance and practicality.