Abstract:

Currently in Yioops discussion board system, various statistical data are calculated such as number of users belonging to a group, number of views of a thread, etc. When statistical data are made publicly available, there is no guarantee that the privacy of an individual is preserved. This project deals with implementing a privacy system for statistics generated by the Yioop search engine and discussion board system. Differential privacy preserves the privacy up to some controllable parameters of individuals when statistics from a database are made public. With this measure, accurate information about the database is provided while at the same time, privacy of the individual is maintained.

CS297 Results

• Gave a presentation on Differential Privacy, understood Yioop system and it's privacy.
• Added statistical chart when clicked on most visited threads in Yioop.
• Developed a test suite of statistical attacks against query and discussion board statistics.
• Implemented differential privacy to number of views of each group's thread.

Proposed Schedule

Key Deliverables:

• Software
  • Enhance Security Feature in the UI by adding a turn on/off button for Differential Privacy under Security section of the Yioop search engine.
  • Randomize user information in the database by obfuscating the user id in order to prevent the leak of user data.
  • Add Differential Privacy to query statistics page that displays the statistics about each query entered by a user in the search.
  • Add Differential Privacy to group statistics page that displays the statistics about each group's number of views.
• Report
  • CS 298 Report
  • CS 298 Presentation

Innovations and Challenges

• Differential privacy is achieved by fuzzifying the actual data. The data is fuzzified by adding some noise to it. The challenge here is to calculate the amount of noise that should to be added to the actual data while still making it as accurate as possible.
• With reference to Deliverable #2, given the condition that Yioop's database is made publicly available, what sensitive information should we consider while protecting the privacy of an individual. For example, if we think a user's id is senstive, we could randomize a user's information by obfuscating the id of the user. This could be done by using hash.
• There are limited papers that are published on the topic of Differential Privacy. So doing research on this topic and implementing differential privacy is a challenge.

References:

Dwork, C. Differential Privacy, 33rd International Colloquium on Automata, Languages and Programming, part II, 2006

Dwork, C. and Roth, A. The Algorithmic Foundations of Differential Privacy, Foundations and Trends in Theoretical Computer Science Vol. 9, Nos. 34 (2014) 211407, 2014