The Automated Ban List - Page 162

On October 24 2010 07:54 Gummy wrote:
Consider the model of TL as a markovian process:

We take vector u to be a duple ut<u1,u2> where u1 is the number of good posters and u2 is the number of bad posters at time t.

Suppose each time step applies a matrix transformation A

[A1 A2]
[A3 A4]

to this vector u with the entries of the vector A listed s.t. the resulting vector after performing Au is given by



This model then presumes that some combination of the following things can happen...

1.) The number of good users/bad users increase/decrease at an exponential rate.
2.) Good users can become bad users, vice versa, at an exponential rate.

These behaviors are consistent with the aims of the site, so I'll propose a concrete example of an A that applies a desired transformation over a given time step.

I will define the following variables,

G = growth rate of TL's poster base.
K = proportion of new posters that are good posters.
P = proportion of bad users permabanned.
T = proportion of bad users temp-banned. (presumed to become good users afterwards)
D = proportion of good posters who defect to the dark side.

The following 2 equations basically say, the number of good users in the next time step is a function of
a.) the number of new good users that have joined (KG) plus
b.) the number of bad posters that have been reformed by temp bans (T) minus
c.) the number of good users that defect to the dark side (D).


The following set of equations says that the number of bad users in the next time step is a function of:
a.) The number of good users that defect to the dark side (D) plus
b.) The number of bad users that join the site ((1-K)G) minus
c.) The number of bad users that are banned permanently or temporarily (P+T)



The resulting equations that summarize the matrix transformation are given by the substitution of the above definitions into the previous equations:


The reason why I'm using matrix multiplication for this is that we can now do diagonal decomposition into eigenvalues to observe equilibrium behavior. (I'll add the calculations for this part later since it's even more nerdy than what I've already listed...)

The point though is that to reduce the number of bad posters at any given time, there are 3 ways:

1.) Give out more bans to bad users.
2.) Decrease the growth rate of the site.
3.) Increase the proportion of new users that are good posters.

Of these options available,
1.) Costs a lot of time and energy. And I don't think there is any way you can really motivate banlings to do a comprehensive purge of, say, the HoN/DotA subforums where garbage posters dwell and multiply.
2.) Is not desirable to the well-being of the site.
3.) Is easy and only takes a constant amount of time. Even if we can change the proportion of new users who are good posters by a few percent, it will be a huge difference over the long haul.

In the end, my argument comes down to the opinionated assertion that updating the 10 commandments will at least increase that proportion of new users that are good posters by a little at the cost of virtually nothing.