Detection And Study Of Spam APP Based On Metadata

In recent years,due to the openness and ease of use of the Android platform,based on the Android smart phones become popular for more and more consumers,therefore,increasing developers turned to Android application development.Some enterprises,groups or individuals in order to promote their own apps,they will upload a number of function similar apps or add some non-related descriptions and keywords about their apps to improve the probability of being found,there are some malicious attackers who add some violating user privacy codes to some popular applications,and then re-uploaded to the application market.As the application market not only lack of effective detection,and lower upload threshold for uploading the app,so criminals use the related vulnerability to upload some spam app.These spam apps not only reduces the user experience,but also increase the burden of application market staff.Spam apps have many obvious features,for example,no specific function,not related app description information,not related keywords,similar functional applications and so on.Application market staff through the manual method in accordance with the spam app identification rules detection app,and remove the apps which are identified as spam,this method is inefficient and cannot found and delete spam app timely.In this paper,a new spam app detection method is designed.At the same time,the results of different classifier models are compared.The precision,accuracy and recall of spam app are all very good.This is the work of this paper and their predecessors' different places.This paper first uses some obvious features as the checkpoints,and collects the data sets used in previous experiments,and then uses the decision tree,support vector machine,random forest and Boosting classifier model to experiment.The decision tree classifier model is widely used for its good readability,human analysis,high efficiency,and repeated construction.SVM has many advantages in solving small sample,non-linear and high-dimensional pattern recognition.The random forest is a classifier that uses multiple trees to train and forecast the samples.The output of the random forest is determined by the number of individual output results.The Boosting algorithm is a method of integrating several classifiers into a classifier.Finally,the experimental results are compared,and the difference of experimental results of different classifiers model is compared,and the reason of the experimental results is analyzed.On the one hand,this paper proves the validity and correctness of the selected features in the paper,and on the other hand,we can use the classifier model to design the spam app.