Android app behaviour classification using topic modeling techniques and outlier detection using app permissions
Mayank Garg, Akshit Monga, Priyank Bhatt, Anuja Arora · 2016
Now-a-Days consumption of Android apps has become a common phenomenon but user switch from one app to other app is also having high expectancy. There are various causes of Apps' swapping by users. As per research study, one prime reason behind this is that android apps are not providing same functionalities as mentioned in their description on Google Play Store and second crucial reason is that Apps accessing users phone content without taking their permission. The objective of this research work is to classify the apps effectively and identify/detect outlier apps with the help of app behavior analysis. Outlier apps have been detected to validate whether an Android app performs as it claims in its description on Google Play Store as well as other criteria is App accessing user's personal content without user's agreement. This work has been done in four phases which are as follows-Data extraction phase-apps content such as App Title and Description has been crawled and extracted from Google Play Store; Data Pre-processing-this preprocessing phase is required to reduce missing data and high dimension data using filtering and stemming techniques; App classification: formed clusters on the basis of generated feature vector list of various category apps with the help of Topic modeling approaches-probabilistic approach LDA and deterministic approach Non-negative matrix factorization approach NMF; Outlier Detection:- finally for outlier detection used manifest file/ user permission file off apps and mapped its content with App specific features list content to find out outlier Apps.