See AdvoGato for an example of a similar "trust" system.
See [Bouillon] for working implementation of similar principle (reputation and opinion-aware peer-to-peer wiki).
[Most of this page was created in early June 2000, before the flurry of AdvoGato-related material in August 2000.]
My vision of large-scale moderation for online forums like UseNet has always been based on the WebOfTrust model. My basic goal is the same as CliffordAdams with his ViewPoint concept: A content-filtering method that tailors itself to each user's needs. --MattBrubeck
This concept is a moderation system similar in function to the one at SlashDot, where posts are rated by readers. Each post has a "score" based on the sum of the ratings it has received, and users can choose (for example) to read only highly-rated posts.
The proposed mechanism, in brief:
- Any user can rate any content.
- A user can declare that they trust someone, to some degree. Through a WebOfTrust, the user then also has some implicit trust in anyone whom that person trusts, etc.
- Each reader sees content ranked based on the ratings of all individuals he or she trusts. Ratings are weighted more heavily if they come from well-trusted individuals.
Benefits:
- Each user can choose what sort of moderation/filtering to receive.
- The moderation system is robust against careless or malicious users.
- Everyone has the ability to moderate. Highly trusted community members serve as editors for the whole community.
Questions:
- How can this be implemented simply enough that people will use it?
- To how small or large a community can this scale?
- How soon can we build it? :-)
"A user can declare that they trust someone" - Slashdot calls this meta-moderation - rating the people who do the rating. One way to make it simpler is to make it implicit. You "trust" the opinions of people who rate the same way that you do.
One way to think of this is as rating-prediction. The system tries to predict what rating you will give an article which you have not yet seen, based on how you rated previous articles and on how others rate them. How you actually rate it then provides some feedback which the system can use to correct its future predictions. The prediction algorithm can become arbitrarily complex.
Amazon used a system like this for predicting which books you will like, with the idea of selling you more books. A message-filtering system could filter based on your predicted ratings.
Further reading on reputation
Blogs
- http://www.smartmobs.com/archives/cat_the_evolution_of_reputation.html
- http://www.peerfear.org/rss/subjects/index-reputation.shtml