Abhishek Choudhary - 2012-10-16

Privacy Preserving Url Recommender is an application that provides url recommendations on the basis of the websites browsed by a user using Collaborative Filtering(recommending items of one user to all users similar to him in their interests).Its architecture is based on the Privacy Preserving Personalisation Architecture(P3) proposed by Bell Labs, Alcatel-Lucent Technologies.
The term privacy-preserving means that nobody ( including me ) can tell which urls have been browsed by you. This is a model that seeks to replace the existing recommendation systems like youtube , wherein you have to expose all your browsing history to Google to get recommendations and personalised services. This architecuture makes it possible to get Youtube recommendations without signing into Google. This software records your browsing history and then provides you the recommendations based on URLs browsed by other users similar to you.Its based on a trust No One model, wherein you dont trust even the recommendation provider(Me in this case).

I guess you must be curious to know how this system works. The key idea of this architecture is to identify similar users without actually comparing items among users( done using Local Sensitivity Hashing(LSH)) and then to push a mashup of items of all such similar users to a particular node in a decentralized middleware. This makes it impossible for the person hosting the middleware to tell from which user this item came from . The data is transmitted over an anonymous channel using TOR , so there's no attack by IP tracking. Though this is the description of the system in a very simplistic way, the actual architecture is far more complex involving 4 layers-
1.TOR Anonymiser
2.Distributed Hash Table

  1. Cassandra distributed Database
    The middleware is hosted on two computers in SIT, IIT Kharagpur. To get a more detailed view of the idea,you can have a look at this http://bit.ly/p3-idea.
 

Last edit: Abhishek Choudhary 2012-10-16