How do you establish trust in a product online, or in a web site?

I’ve been spending some time looking at online reputation systems recently. I’m in the information gathering stage of my research, and I’m finding some interesting articles on the subject.

Looking around the web, it’s not unusual to see some reputation systems in place on sites such as eBay, where buyers and sellings can provide feedback on transactions they’ve been involved in. Or in the customer reviews on Amazon.com. Some online forums include reputation systems for posts and posters.

One of the conclusions I’ve come to is that an excellent way to test an online reputation system is to see how difficult it might be to attack that system, and see where it might be prone to fail.

An article I found took a close look at Google’s use of PageRank as an online reputation system, and considered methods of attacking it to see how effective a system it is. Andrew Clausen’s Online Reputation Systems: The Cost of Attack of PageRank (PDF) is an interesting look at the ranking mechanism used by the search engine to gauge how popular a site is based upon the value of links to the site.

The paper has been out for a little over a year, and it really hasn’t attracted much attention. There’s a lot of math in the middle that might make it reading suited for a fairly small audience, and there are some statements in it that I’m not sure I agree with. But, if you can skim past some of the math, it does make for interesting reading.

This is an early quote from the paper that drew me in, and had me reading through most of it (except for some of the calculations):

Perhaps the best way to evaluate a reputation system is to measure the cost of attacking it. That is, measuring the cost of increasing an individual’s reputation through deception. If this cost is high, then traders can be confident that reputation scores were obtained legitimately. Moreover, they can assess whether a trader’s reputation is sufficiently high to be trusted. Clearly, a deceptive trader may well pay to attack a reputation system if the payoff from a trade were large enough.

Therefore, an important goal of reputation systems research is to design a system that has a high cost of attack. I argue that Google’s PageRank algorithm is a promising foundation for reputation systems, since all information is weighted by the reputation of the source. Therefore, the aim of this thesis is to determine the cost of attack of PageRank.

It’s worth reading for the comments in the paper on “link farms”, SearchKing, and the failure of Google to include a complaint aspect in their reputation system.

I didn’t expect to uncover an article about Google while scouring the web for information on online reputation systems. I’m glad I did.