Re: creating a decentralized web of trust

On 26 July 2015 at 19:59, mike amundsen <mamund@yahoo.com> wrote:

> "trust" is a belief on an unbounded scale that changes over time (both the
> scale and the belief change).  the assessment of trust is ALWAYS local
> (e.g. between you and me) but MAY be shared (e.g. i tell my friends about
> my assessment of you).
>
> when sharing the assessment w/ someone (or a group), the value I use is
> ALWAYS translated to the "local" value of my share target. e.g. when I tell
> my mother how much i trust you, my mother translates my trust assessment of
> you to HER scale to generate HER assessment value -- this continues for
> each person with which i share the assessment, even when i share it with a
> large group like a meeting or concert hall.
>
> successful trust models allow us to use our own assessment scale
> internally and translate others' trust assessments to our own scale when
> they are shared with us. this is not *easy* but is also is not
> *complicated* -- it's complex.
>
> AN EXAMPLE
> in the U.S. the "credit score" is an example of a system that handles the
> negotiation of assessments. assessments from many places are translated
> into a local scale (for that reporting company) and each translated
> assessment is used to create a single value (your credit score) and *that*
> single value is used by others to make their *own* decision about your
> credit-worthiness for some single event (getting a credit card, buying an
> automobile, etc.). it's worth noting that, the U.S. there is more than one
> credit scoring company and they all have their own scale/model AND some who
> are making credit decision use multiple reporting companies as inputs when
> making their own assessments.
>
> it's turtles all the way down ;)
>

Thanks, this makes total sense.  I'm hoping not to make something perfect,
but maybe "good enough" to help with some use cases.  As we start to see
more writing on the web, I think it's going to be important to spot
spammers, so this is one thing I'd like to try and reduce if possible by
verifying IDs as having some kind of footprint.

Will think more about how to get the "local" value of a trust metric.


>
> cheers.
>
>
>
>
>
>
>
>
> mamund
> +1.859.757.1449
> skype: mca.amundsen
> http://amundsen.com/blog/
> http://twitter.com/mamund
> https://github.com/mamund
> http://linkedin.com/in/mamund
>
> On Sun, Jul 26, 2015 at 11:54 AM, Kjetil Kjernsmo <kjetil@kjernsmo.net>
> wrote:
>
>> On Saturday 25. July 2015 21.34.24 Melvin Carvalho wrote:
>> > I multiply the #followers * 3 up to a maximum of 30 followers.  e.g.
>> >
>> > http://gitpay.org/torvalds -- 90%
>> > http://gitpay.org/stratus -- 9 followers = 27%
>> >
>> > I am looking for ideas on how to improve this algorithm, or maybe find a
>> > set of algorithms people can choose from to get out a trust score
>> > (however i am scpetical people will have time to code them).
>>
>>
>> Very cool already!
>>
>> To convert numbers in an interval from 0 to something unbounded into
>> something, I tend to suggest functions that will do it, rather than cap.
>> For example
>> 1-e^(-x)
>> That will always make sure that the number you get back is in the [0,1)
>> intervall, if the input is in the [0,Inf] interval.
>>
>> And then, you can add some factors to adjust it. The first thing to do is
>> that you might like it to be in the [0,100) interval instead, and then you
>> just do
>> 100*(1-e^(-x))
>> and then, you probably want to adjust how fast the score goes up, to do
>> that, you can introduce a factor r, like this:
>> 100*(1-e^(-x/r))
>> You can then adjust r based on your assumptions of what a "large x" would
>> be. So, following your example, where 30 is a large number of followers
>> that should result in 90%, you'd have r = 15.
>>
>> You could fire up R (see https://www.r-project.org/ ) to explore the
>> function, you could define it like this:
>> score <- function(x, raise) {
>>   return (100*(1-exp(-x/raise)))
>> }
>>
>> and then create an array with integers from 0 to 30:
>> x <- seq(0,30)
>> and then plot:
>> plot(x,score(x, 15), type='l')
>>
>> But to use something more than just #followers for x, perhaps PageRank
>> would be the modern way to compute it these days...? :-)
>>
>> Cheers,
>>
>> Kjetil
>>
>>
>