Tag Archive for 'trust metrics'

Trust metrics are techniques for predicting how much a certain user can be trusted by the other users.CouchSurfing doesn’t really have a prediction mechanism, but trust values are registered for every friendship link.

I never thought the denominators for the trust value made a lot of sense for the friendship links on CS (especially when translated, I don’t really know how to best translate “I somewhat trust this person” into my mother tongue). Still, there seems to be a definite trend of linearly (in time) decreasing trust on the Quality of Service page. It would be interesting to compare this to values from before and do a deeper analysis. The “average quality” doesn’t seem to be changing significantly on the other hand, maybe slightly going up? Possibly because it’s actually visible to the receiver.

Joe Edelman wrote the QoS code, and wrote to me:

Wow, that *is* interesting!

So the avg trust is calculated among introductions added in the last
week that are reported as due to CS and in-person. So it’s not because
of virtual users, and it’s not because CS is accelerating and includes
less pre-existing friends.

The only confounding factor I can think of, is that it doesn’t take the
“date you met this person” field into account — a lot of people don’t
fill it out, or don’t fill it out correctly. So it includes
introductions that are finally being reported from the past, as well as
those that actually occurred that week.

We could be seeing an ever-greater percentage of weirdos from the past.
You know, those random people that blew through a collective, and much
later are friending everyone. And the people they are friending hardly
remember them and so don’t trust them. This would be a result of social
graph “fill-in”, perhaps as a kind of recoil from expansion last summer.

Or, perhaps it’s an accurate result, and as CS grows, people that meet
find they have less in common, since CS includes more demographics.

In that case, it could be interpretted as a *positive* result: perhaps
the ideal would be to take people who DON’T trust each other INITIALLY,
and give them POSITIVE EXPERIENCES such that later they DO trust each
other, or they start to trust other people from a new demographic MORE.

Let’s just hope this trend does not continue. If it would, the average trust would be zero by the end of 2010.

In the meanwhile, some active work can be done on designing and implementing a trust system from scratch on BeWelcome.

The data:

year   week introductions users  quality  trust

2008    16         6625   3890    1.526   0.370

2008    15        14238   7345    1.506   0.377

2008    14        14818   7591    1.490   0.379

2008    13        16520   8201    1.527   0.388

2008    12        13895   6952    1.500   0.387

2008    11        12252   6291    1.479   0.379

2008    10        12303   6490    1.493   0.392

2008    09        12796   6482    1.480   0.382

2008    08        11336   5875    1.483   0.376

2008    07        12484   6408    1.486   0.391

2008    06        11778   6215    1.469   0.409

2008    05        11201   5945    1.453   0.406

2008    04        10570   5998    1.479   0.415

2008    03        10757   5983    1.489   0.410

2008    02         9560   4872    1.503   0.410

2008    01        13972   6425    1.484   0.417

2007    52         7749   4279    1.476   0.414

2007    51         9332   5118    1.467   0.421

2007    50        10975   5500    1.480   0.422

2007    49        10309   5632    1.454   0.415

2007    48        10664   5500    1.454   0.413

2007    47        10335   5734    1.487   0.425

2007    46        10835   5762    1.492   0.429