Beyond recognition

Posted on 19 Jan 2007 at 16:31

 

Gallery

A further test, taking in 100 users, showed that the number of sessions required for a usably accurate identification (one accurate over 90 per cent of the time) rose to between just 18 and 20. This is still low, especially when you consider that the algorithm used just five general variables as its inputs. More detail should result in greater accuracy from fewer sessions.

There could be a number of practical applications. Supposing, for instance, you use a supermarket's online delivery service to save you the trouble of lugging heavy tins, packets and jars back from the shops. After just a few uses of the service, the site might begin to recognise you and configure itself according to your tastes even as you navigate to the login screen. But one day 'you' go immediately to the vintage champagne and fresh strawberry sections and spend a fortune. Is it you, or someone wining and dining at your expense? By realising that this activity doesn't conform to your clickprint for that site, the online shop may suspect something and alert a customer service representative, who could call you directly before allowing any payment authorisation.

Writing analysis

There are less-welcome implications to clickprints, though. Regardless of your browser's privacy settings, online marketers might be able to identify you and bombard you with targeted adverts. This would be done without you logging in or registering, by exploiting your unique site usage profile. Even using a proxy server, which hides your true IP address behind its own, would be useless, because how you surf always gives you away. What's to stop someone impersonating your surfing style? They might then leave messages or even spam in your name on an online forum or as a blog comment.

A team in Arizona says they have a technique that can help prevent that and also ensure that the person you're talking to online really is who they say they are. Ahmed Abbasi and Hsinchun Chen from the Department of Information Systems at the University of Arizona have come up with the concept of 'writeprints'. Similar to clickprints, these also serve as online fingerprints. However, this time a person is identified from entries on blogs, Usenet, online forums and even chatrooms.

Cyber-house rules

The pair were concerned that cyberspace allows less-welcome visitors to flourish undetected by passing themselves off as respectable users, often with the support of other 'users', who are in fact the same person using a different online name. "In addition to using the internet as an illegal sales and communication medium, there are several trust-related issues in online communities that have surfaced as a result of online anonymity," said Abbasi. These issues include the problem of accurately identifying strangers you or your children might meet online and with whom you may strike up a friendship. In the real world, you can see who someone is, but online you have only words to go by. If there were a way to check that person's other posts on other forums, you could spot any anomalies and know all was not what it seemed.

According to Abbasi and Chen, this is now possible. Over time, your writing style expresses itself subtly yet as clearly and uniquely as a fingerprint does to a forensic scientist. Abbasi and Chen have devised a way of analysing your writing style and so identifying the true authors of messages, even when they have used multiple accounts to cover their identities or to vouch for one another. The writeprint identification system created by Abbasi and Chen is grounded in a field of statistics called stylometry. This is the analysis of writing style based on what is written, not to be confused with graphology, which is the analysis of the shapes made when someone puts words on paper.