Privacy Implications of Not-So-Hidden Comments in arXiv Files and Analysis of Online Privacy Policies

Abstract

Internet users can accidentally expose their own private information in a myriad of ways. This paper describes our approach to a large-scale measurement study on one case of online privacy leakage wherein users upload files for publication and sharing, files that can contain users’ private information hidden within them. We analyze comments in TeX source files of arXiv publications using various natural language processing techniques to identify specific attributes of comments that may represent privacy violations. We also perform near-duplicate detection and clustering on a large data set of privacy policy texts to understand how online privacy policy is communicated to users. We find that arXiv publications contain many interesting comments despite the ease with which authors can strip out all comments. We find that many privacy policy texts are duplicates or near-duplicates of one another.