My Idea on the Yo-Yo Effect

[jlepine]

The evolution of TR (trust rank) would be proportional to the difference between the TR at a time t and the value of the keyword in question W. (Our law Newton us what!)
Equation: dTR / dt = k (TR - W)
We know the solution TR(t) = A*exp(kt) + W ....
It remains to use the conditions which lead to ... It is no doubt take TR(0) = 0 which gives the
It is understood of course that this function can only be the average function of the evolution of sites in relation to a given keyword!

TR(t) =-W*exp(kt) + W

This function rises significantly towards the end and is very flat for a long time.
This causes a flat classification that passes easily from the place 8 to 17 for example ...
While in the growing once it reached position 5 will lose many points and down!

[tedster]

I really don't understand your idea well enough to comment - sorry.

[Shaddows]

I disagree.

Fundamentally, TR does not (necessarily) vary with time. I could stop there, but...

While RANKING (not TR) might vary with time, you cannot easily translate any given variable score into any rank, due to the amount of unknowns. The pertinant unknowns here are:
1) The other 200-400 variables establishing a rank
2) The TR of competing sites

Chosing TR as the variable controlling the Yo-Yo is arbitrary at best. In fact, I would say it could be categorically ruled out.

In any case, the Yo-Yo only seems to affect 2-4 sites in a given keyword SERPs. If the Yo-Yo was controlled by a (relatively) simple function, all sites should be bouncing around in a chaotic but deterministic way. This applies to whatever variable you posit to control the Yo-Yo.

Finally, and forgive my rusty calculus, but your final equation can be rearranged to
TR(t)= W[1-e(kt)]

The striking feature of which is that Trust Rank is defined entirely by some product of the keword value. Defining any attribute of a site in terms of a keyword it competes for is nonsensical.

The other interpretation is you meant to write
TR(t) =-W^(kt) + W
"exp" becomes "^" and reads "to the power"

In which case trust (or indeed any subject variable) becomes negative very quickly, or the constant is negative (TR defined almost entirely by KW value), or between 0 and 1, in which case TR suddenly become negative once t>1/k

[edited by: Shaddows at 10:02 am (utc) on June 12, 2009]

[jlepine]

you're right
the index of which I speak, and I call on my IG site is necessarily linked to the "keyword" in question.
When I talk about TR is the index related to the keyword I'm talking about!
And of course, the function must be considered as an average evolution of the indices of all sites linked to the keyword in question.