« Newer Is Not Always Better | Main | Reputation Can't Be Asserted »
Some Thinking About Reputation
In my grad class this semester, we’re designing and building a reputation system. Today we had some discussions which I wanted to capture and get feedback on. First, the overall idea is that reputation is computed from identity and transactional data. So a reputation, R, is calculated as follows:
I == a vector of identities
TxI == a vector of transactions on I
VI == a vector of verification data on I
R = F(I, VI, TxI)
Some thoughts
- Allow users to assert I
- The system would provide ways for users and others to verify I (forming VI)
- The system collects related transactions for identities are agreed to by users and provided by relying parties and others.
- Relying parties would authenticate to gain access to the system.
- The function F could be set or changeable. Our thought is that each relying party would be able to define their own F based on data in the system.
- R would be available to relying parties and the user.
- Transparency is important, so users should be able to tell why they got the score they did and see how they could improve the score.
Posted by windley on February 3, 2006 11:19 AM
Comment from Dennis Wood at February 4, 2006 11:12 AM
Phil,
Obviously I'm not a programmer, so speaking purely theoretically it would seem to me that the formula should include:
(1) a vector of identities
(2) a vector of transactions on I
(3) a vector of verification data on I AND T
Dennis
Comment from Phil Windley at February 4, 2006 11:36 AM
Yes, that's possible. I'm not sure how often transactions will be verified by some other source. But it certainly doesn't hurt to incldue it in the data model for that possibility.
The idea behind verifications on I are things like the common scenario where you assert an email address as an identifier and then a verification message is sent to that email address. When it's clicked, the email is "verified."
Other example include placing tokens on Web sites that you claim or even your physical address (with postcard verification).
I should probably be split into I and A_I, a set of attributes associated with a given identifier. Both I and A_I could be verified in many cases.