Research On Terminal Privacy Data Protection Technology Based On Contextual Integrity And Multi-label Classification Technology

With the in-depth use of mobile terminals,people keep more and more personal information on their mobile phones.The Android system is currently the most popular platform for mobile terminals.There are tens of thousands of applications developed based on Android.While providing personalized services,these applications are also continuously collecting data on mobile terminals to provide more detailed services.However,Android's own permission system cannot effectively evaluate the data acquisition behavior of the application,which has threatened the security of personal privacy information stored on the terminal by Android users,and the application collects and uses personal information without the user's consent.Has become a recognized violation.In view of the above problems,this thesis proposes a new runtime privacy data protection system combining modern privacy protection theory and Android runtime behavior information.This article mainly includes the following research contents:(1)From the perspective of privacy data dissemination,this article combines professional privacy protection theory to conduct a risk assessment of the application's privacy data calls.In the past,privacy protection systems learned more about users' privacy intentions,and privacy intentions were susceptible to user subjective factors.This thesis uses the contextual integrity theory and the Android permission model to propose a context integrity calculation model for the Android platform.This calculation model can quantify the risk of privacy data invocation and avoid subjective factors in the user's intentions.This improves the accuracy of protecting private data.(2)In order to support the privacy data protection system based on the context integrity theory,this thesis uses the runtime behavior information and context information to train a multi-label classification model,which can identify the scene topics contained in the context information of the Android runtime.During the model training process,the base model of word frequency features and topic features are designed respectively,and a model with better performance is used to obtain a secondary model with better performance as a privacy decision support model.In this thesis,the best algorithm and optimal parameters of the training model are selected through experiments.(3)Based on the theory of contextual integrity,this thesis designs a runtime privacy data protection system.This system is developed based on the Xposed framework and can protect privacy data in real time during runtime.The system is divided into offline privacy decision aided model training and real-time privacy data protection.This thesis designs experiments to verify the effectiveness of the system in privacy protection and the usability of the privacy protection system in the terminal.