okay so here’s what you do
make a row matrix with.. say, 50,000 cells. Each cell will represent a potential topic. For this simple model all topics will be equally likely but you could tweak this.
then
imagine a world that proceeds with discrete periods. that is, t=0,1,2,3,...
the value in each cell in a given period is denoted V(cell, t), where cell is a whole number in the set [1, 50000]. I claim that V(cell, 0)=0 for every cell; that is, each cell starts out with the same zero value.
For t > 0, we let V(cell, t) = p(V(cell, t-1) + (1-p)ɛ where ɛ is a stochastic variable from a distribution I’ve yet to decide; it is determined in every time period. p is a constant value in the area [0, 1] that could differ for each cell – this would be interesting, as in this case p would be the inertia of a topic. That is a higher p would result in a topic more difficult to start but more enduring once it became prominent.
As I think more about it this is not quit desireable – we should not necessarily link the stochastic factor with the lag factor so explicitly.
Hmm. This is a start but it’s not quite swinging enough. I’m going to think about how to cause burstiness in these threads but I think it should be pretty straightforward.
Back to work!
Post updated by sam on 2008/06/20 11:11
