Search Engine Optimization, Blog Design, Movable Type Customization, and more.
Web Dawn

P2P REPUTATION SYSTEMS



My sponsered child, Hama from Niger, Africa
A portion of the proceeds from this site help sponsor Hama from Niger. Click here for more information.





Subscribe to the RSS feed for this conversation to monitor future messages View previous topic :: View next topic  
Author Message
Mark







PostPosted: December 31, 2003 09:18 AM 

Can peer-to-peer (P2P) reputation systems work?

Some of the successful and well known reputation systems on the web are controlled and managed by a central entity. An example that I have used in the past is the eBay feedback system. eBay enables buyers and sellers to rate other buyers and sellers, and presents that reputation information to everyone. Another example is the industry for digital certificates in which there are a few large, trusted "authorities" who vouch for the reputation of the certificate holder. In both of these examples, the reputation system is controlled by a large corporation - giving that company a lot of power, competitive leverage, etc.

Since power in the hands of large corporation can easily corrupt, the idea of a peer-to-peer reputation system - one that is not controlled by a single entity - is very appealing to many (including myself). But can such a system work? And if so, how?

First, I think we need to look at the main reasons why centralized reputation systems are successful. One of the most important reasons is trust. These systems work because people trust the reputation information presented by them. When look at the eBay reputation of someone, I have a high degree of confidence that the person's feedback score is accurate - I don't stop to wonder if that person has hacked the eBay system and changed the score. The eBay system works a s a reputation system because people trust eBay.

But in a P2P approach, there is no large, recognizeable organization. Reputation information would be hosted in many different places, perhaps on the web sites or computers of individuals whom you do not know. So how do you know that you can trust these people? In order to trust them, you need to learn about their reputation.....here the problem becomes circular in nature. Can can we be confident that reputation information information in a P2P system is not forged or manipulated? One solution that comes to mind quiclky is that the reputation information could be stored in more than one place - in which case an anomoly could indicate attempts to forge data. But what happens when reputations legitmately change - the change must propagate to other nodes - how do you know that it is a real change and not a forged change? How can we trust reputation information in a peer to peer system?

For the answer, I think we need to look at the most common reputation system of them all, one that has been around for thousands of years: word of mouth. For moment, let us forget about the web, TV, radio, newspapers - without these sources of information, how do we learn about the reputation of others? Personal interaction is the first way we can do this. If we buy products from a vendor at a (social) market, based on the quality of the products and the vendor's claims, we develop and opinion about the reputation of that person. Another way is by talking to others, sharing our opinions about the reputations of others. If I believe that the market vendor has a good reputation, and John believes the same, it reinforces my opinion of the vendor's reputation. But what if Mary tells a story about the vendor in which she was cheated? That might decrease my opinion of the vendor. This also works for people we have never met. Before I walk into a new shop, I might ask a few people about their opinions and experiences with the shop owner. Some of the people I ask may not have been in the shop either, but may relate stories that others have told them. This way I gather reputation information for someone that I never met - which is exactly what we need to do on the web! but what happens if I get 10 conflicting opinions about the shop owner? How do I use these to form my own opinion? The answer again is trust. Some opinions I will weight heavier than others, depending on how much I trust the person giving the opinion - depending on their reputation. I trust the opinion of my best friend more than that of a casual aquaintance. This raises an important difference between a centralized reputation system and a word-of-mouth system: with a word-of-mouth reputation system, reputation is relative, not absolute. In the real world, a person's reputation is seen throught the eye of the beholder. Can we harness this word of mouth system on the Internet?

The word-of-mouth system, as described above relies on a social network to gather and disseminate reputation information. Online social networks can be used in the same way, whether they are centralized systems or peer-to-peer. Now we can revisit the question: how can we trust reputation information in a peer to peer system? The answer is that you trust the people that you know best, just like in the word-of-mouth system. Online social networks already track the friend-of-a-friend and six degrees of seperation, and P2P applications can do the same. Members just need to input the reputation information for the people that they know directly, and reputation information could aggregated and persoanlized based on each person's trust level and degrees of separation. This way you don't need a big corporation to "trust", you just need to trust your friends. Perhaps some of the online social networks are already moving in this direction, I am not sure. But it seems to me that the answer is "Yes", P2P reputations systems can work.

Prashant Dewan







PostPosted: January 9, 2004 12:28 AM 

Mark

I agree with most of your comments but there are more fundamental problems to a decentralized p2p reputation system which has been by and large ignored till now.

How do you identify the peers to which you assign the reputation? Can peer have one identity or can it have multiple identities? Should the peer be traceable from one or more of its identities? Should it be possible to find out the other identities of the peer , if one knows one of the identities of the peer?

There is a long list of questions, I do not have very good solution to any of them . I do have some ad hoc solutions though, which work in "most" cases. Its a battle yet to be won!

The solutions are on my website.

lily







PostPosted: February 10, 2004 09:18 PM 

Is it necessary that in a p2p network every peer need value reputations according a uinform standard?
How can we do?

paolo







PostPosted: April 9, 2004 12:11 PM 

Hi, very interesting post.
Actually I'm studying (phd) these topics. And I agree with your visions at all!
I want to point you to:

- This project "Trust and Reputation in Web Based Social Networks"
http://trust.mindswap.org/

- This paper "Trust Networks on the Semantic Web"
http://www.mindswap.org/papers/Trust.pdf

- My (messy) wiki "trust metrics evaluations wiki"
http://moloko.itc.it/trustmetricswiki/moin.cgi/FrontPage
(feel free to edit it)

By the way I prefer to use trust (that is more subjective) than reputation (that seems the global averaged value of someone in a community).

In my vocabulary, local trust metrics compute what you call "relative" (personalized) reputations (or trust values). and Global trust metrics (PageRank for instance) compute "absolute" trust values.

Comments on comments:
Identification: if you want something decentralized the only possible way is to use public keys.

Standard for expressing reputation:
There is an extension of FOAF for doing this
http://trust.mindswap.org/trustOnt.shtml .
the max trust is 10, the min is 0.

That's all, I'll keep reading your blog.

Mark Carey







PostPosted: April 21, 2004 09:06 PM 

Thanks for the link, Paolo.

Actually I came across that site recently -- I am pretty sure I found it via your blog, so thanks again. :) I find it very interesting, and I really need to spend some more time looking at at -- the concept is very promising.

I also need to read more of your blog. I like your focus on both Trust and recommender systems. I believe this will be a combination that can truly define the social web in the years ahead.

Vishal Sadana







PostPosted: March 10, 2005 05:12 PM 

The problem is really very complex. What I believe is that each peer should have a trust value.This trust vale should be developed on the basis of its past interactions with other peers and the experiences of these peers.Whether like the quality of service being provided,has a particular peer behaved properly in the past.
Once the trust value is attached to every peer,the next goal is how to propagate this value in a secure manner thorughout the network.Assuming that there is a significant number of malicious peers, I am concerned with the protection of the invidual trust value.

Jim Beggs







PostPosted: August 2, 2005 05:25 AM 

Re trust & repution in online networks, we have just launched www.biddingBuddies.com, a social & reputation network exclusively for eBay members.

Jim.

VISHAL SADANA







PostPosted: August 28, 2005 02:18 AM 

I agree with the above mentioned comments.Each peer should have a trust value.This trust value of a peer should be developed on the basis of its past interactions with other peers and the experiences of these peers with the peer in question. In other words, it depends upon the Quality of Service provided.
Once the trust value is attached to every peer,the next goal is how to propagate this value in a secure manner thorugh the network.Assuming that there are a significant number of malicious peers, my main concern is the protection of these trust values while propogation.

Lori







PostPosted: October 1, 2005 02:16 AM 

This is a great post, and though it is quite old I know of a tool that helps to solidify your WOM. It is called iKarma and it allows businesses small and large to have an active role in building and preserving their reputation. It is like eBay for business in a way, and we all know eBay works!

Since we also all know WOM is the best form of advertising and can at the same time cause a companies demise, the tool provides and easy way to collect and store clients comments about you and your business.

By having a third party collect and store those comments and making them available for others to see, a business can gain the value they deserve from the good WOM they have earned. So YES, P2P networking can be managed, and does work!

I was just at the WOMMA conference in New York, and so much on this subject, 'your reputation' was discussed. I really think that everyone should use iKarma - it is a great way to manage your valuble business reputation.

Check out my iKarma if you like - it really works for me!

http://lori.ikarma.com/

Victor







PostPosted: February 3, 2006 02:17 AM 

One more spam link :)
Bouillon P2P universal aggregator project:
http://bouillon.math.usu.ru/
(employs reputation distance as a key criteria of predicted message relevance)

cashmere







PostPosted: March 11, 2006 02:31 AM 

Very good site, congratulations!

Stephan







PostPosted: June 12, 2006 12:47 AM 

Oh, my world. It is ok

sportsbook source code







PostPosted: June 24, 2006 10:42 AM 

sportsbook source code [url=http://nfl-sportsbook.hotultradir.com/sportsbook-source-code.html]sportsbook source code[/url]

inn







PostPosted: June 29, 2006 07:58 PM 

Very good site, congratulations!


Join the conversation:









Remember personal info?

Very Happy Smile Sad Surprised
Shocked Confused Cool Laughing
Mad Razz Embarassed Crying or Very Sad
Evil or Very Mad Twisted Evil Rolling Eyes Wink
Powered by MTSmileys



Check to Subscribe to this Comment:
(email field must be filled in)



Subscribe Without Commenting



Subscribe Without Commenting