On 7/17/13 5:33 PM, Hussein Abbass wrote:
This is more like a controlled experiment: X received information L
then decided to disconnect Y; L is the cause, disconnecting Y is the
effect.
How about:
1) Form a long bit vector of potential Likes. The top 1 million popular
Likes, say.
2) Derive an numeric vector from these by weighting each field by a
vector W. Set W to 1 to start with.
3) Form a matrix of the vectors (the Like adoption per user).
4) Compute a distance matrix of those vectors
5) Perform hierarchical clustering on that matrix
5) For each pair of people, subtract the geodesic distance in the
cluster tree from the actual social network. Square those and sum
across all the people.
6) Use conjugate gradient, or some other optimizer to find a `best' W,
with #5 being the minimization criterion. Add a penalty term for
non-zeros to get rid of uninteresting Likes.
7) Treat the subspace identified (a relatively small number of W
non-zero terms) as a `Like' genome for each person.
8) Use maximum likelihood phylogenetic techniques to infer the evolution
of the Likes. This would involve finding transition probabilities
between Like/Like, Like/Non-Like, Non-Like/Like, and Non-Like/Non-Like,
and overall volatility of individual memes. I suppose there's no real
transition to Non-Like in practice with Facebook, but rather incremental
adoption of more Likes?
The point would be to find the Likes in the genome that best explain
topology changes in the tree.
That is, what memes are associated with cliques being maintained (tree
distance being relatively small) and which do not.
Ok, it's a little cynical to say that proliferation of Likes and Friends
is like the proliferation of a virus, but..
Marcus
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com