Quantification of PIR protocols privacy

In current era the best way to find the information over the internet is search engine. Web search engines maintains user profile for better search results which could raise serious privacy issues. In order to intact the users privacy in front of a web search engines Private information retrieval (PIR) protocols are used which hide the identity of the user by submitting his/her query through other group member. A basic problem is related with these protocols are their predictability. This paper is the extension of previous work in which a person with anonymous query was successfully identified. This paper aims to find all queries submitted by the target user using UPIR and UUP protocols. For experimentation purpose a machine learning based adversarial model is proposed to find the actual queries of user of interest based on the previous profile. The results shows that the precision, recall and f-measure of J48 in finding user's real queries is more then 0.70 on the average. Similarly J48 reported highest trues positive rate of above 0.7 and lowest false positive rate of 0.006. It was also observed that the size of training data has very little effect on accuracy according to this experiment.