Ian Clarke wrote:
> I think our core assumption is that there will be clustering in the 
> graph, meaning that if A is connected to B, and B is connected to C, 
> there is a higher probability that A is connected to C than that A is 
> connected to any other randomly selected node in the network.
> 
> I think given the nature of human relationships, or even the way that 
> people exchange references on #freenet-refs, that this assumption will 
> be valid.
> 
> We should therefore simulate using a graph generated in a manner as 
> close to how darknets are likely to grow and spread as possible.

Agreed - that was the thinking behind choosing the second model, which 
combines preferential attachment with clustering. I'm still not entirely 
happy with it - triads aren't the only form of clustering that's 
important, as Kleinberg's work shows - but it's the best generative 
model I've found so far.

Cheers,
Michael

Reply via email to