Personalized Privacy Risk Assessment Based on Deep Neural Network for Image Sharing on Social Networks

With the extensive usage of social networks, many users get used to share images with their friends frequently without thinking carefully about the private information in the images, which may cause the leakage of user private information. To help users to improve their privacy awareness, in this paper, we propose a two-stage personalized privacy risk assessment framework based on images sharing history between users. In the first stage, the privacy information of the shared images are identified by using the faster R-CNN, based on which, the user-image privacy vector is generated. In the second stage, we predict the user behavior for image sharing using deep neural network and calculate the risk leakage probability of user privacy for the sharing image. The experimental results show the effectiveness of the proposed method, which can reach a prediction accuracy of 98% and the average time of 0.47 s.