The Research Of Privacy Policy Predic- Tion Technology For Text Information On Social Network

Privacy issues of personal data are getting tricky due to the increasing volume of personal data users sharing through online social networks(OSNs).Existing privacy policy mechanisms of OSNs may not protect the privacy of personal data effectively,since users are struggle to set up the privacy settings.To help users set up and manage privacy policies,the concept of privacy policy automation was proposed.The related works mainly focus on privacy policy prediction of photos,locations,profiles et al,few have focused on predicting the privacy policies of text-based posts.The text-based information contains users' behavior or personal opinions is considered to be particularly risky from the privacy perspective.This paper focuses on the privacy policy prediction of the text-based post.Our work is as follows:(1)We proposed a personalized privacy policy prediction model for textual posts.The model takes user's past decisions and textual content as input and outputs appropriate privacy policies to them.Experimental results show that this method's correct predicted rate is over 70%,which can effectively reduce the burden for manually set the privacy policy.The drawback of this method is that the prediction model relies on historical privacy preferences of the user and it cannot resolve the cold start problem.(2)A crowd-based and coarse-grained privacy policy prediction model is proposed to “recommend” the policies for new users.Firstly,find out k posts that are most similar to target post and build crowdsourcing data for training.Secondly,we analyze the textual semantics,emotion and social context of crowdsourcing data to make a coarse-grained privacy policy prediction for new users.Experimental results show that the accuracy of this method is more than 50%,which is better than random guess or select default privacy policy.