Proposed trust system
From BeVolunteer Wiki
This page is a draft. This is tentative, transient, and collaborative.
We the community are in the process of building this.
Please add what you think {{{1}}}, and rearrange things as
you see fit, but don't be quick to delete other people's ideas.
Here is a proposal for a trust system in BeWelcome. Right now the only system that we implemented to give a member an idea about the credibility of some other member is the comment system. That is good but not sufficient.
We want to propose a system that contains ideas about trust that have been implemented and/or suggested elsewhere.
Rationale:
- improving safety
- attracting new members
The concept is based on the following:
- automating normal member behavior - when you judge an unknown person you often check for common friends and their judgment.
- people are different, some get along with each other better than others
- people are different, what is important to you might be irrelevant for me
- public "friendship" statements tend to put social pressure on the people and often end up in false or too positive statements
- the less I know somebody the less likely it is that his opinion about somebody else matches mine
- can be statistically true, but very often it is different! -- lemon-head
- to be useful a system must be extremely easy to use and take almost no extra time for maintainance
- or let's say, to maximize usage
Short description: A member is able to "judge about" other he knows (in person or not). This information is visible only to the member. When the member (visitor) visits the profile of an unknown member he gets a simple visible clue telling him about [ trust | will get along? | credability of this information ]. This information calculated based on the information the member, his friends, friendsoffriends, ... entered. It will be different for every visitor of the profile and not known to anybody except the visitor.
More details:
Add links/data: From an easily accessible place (for example pop up when you click next to a members name) you enter the following information into a convenient way (sliders for variable data, checkboxes)
- if you know the member in person [y/n]
- the time you know (spent with him) him
- your trust level [values from 1-10 on a slider, so similar]
- how well you got along with each other
One could think of one or two more values, but we have to keep it simple.
Contents |
Preferences
In your preference you can define the relevancy of these settings to you (with a good, to be defined default):
- met in person? do you only, want to trust judgments from people who met in person?
- time, is meeting for an hour enough for a good judgment
- suggestion: slider with log scaling, e.g. between minute and hour, between hour and day, between day and week, between week and month, between month and year
- distance. Do you think the judgment of people who are 5 hops away will match yours? or would you only trust those who are 3 hops away?
Calculating the data:
- Get all connections between you (A) and the other member (X).
- Only the shortest possible connections are used (if you have a direct common friend possible connections via 7 other people are omitted). A->B->X and A->C->X but not A->B->N->X
- Only connections that have a certain "trust level" are used. If I don't trust B it does not make sense to ask him about his opinion on X.
- The data is weighted by my preference set, the distance and the rating for each hop
(see also http://www.trustlet.org/wiki/Moletrust)
An example (only calculating the trust level and quality, omitting the "how do we get along"/time/... for simplicity: Me (A) looking at profile of (X)
The following lines connect me A and X with a maximum of 3 steps:
A > B > E > X A > C > E > X A > D > F > X A > G [no trust] H > X ---> this is omitted because G does not trust H
B = Ba (Trust from A (me)) C = Ca D = Da E = Ba(Eb) + Ca(Ec) (Trust from B to E combined with my trust to B and trust form C to E added) F = Da(Fa) X = E(Xe) + F(Xf) == (Ba(Eb) + Ca(Ec) * (Xe)) + (Da(Fa)*Xf))
This might sound complicate and absurd - but as the user will only see some nice catchy things it should not be a problem.
Of course it will take a while until you get reliable results for several members and you will most likely alway don't get a result for most (calculating over more than 5 degrees doesn't make a lot of sense and is resource intensive - so you will probably always only be able to find a link to part of the members).
Some more ideas
- First do a limited test among volunteers before launching it to the whole community.
- Give the user an overview of trust ratings given by that user, so that the user can compare and her values will make more sense.
