User:Agtfjott/Article quality and user creditability/Sandbox

Temporary work area

Mathematical model of the article quality

There are several possible variations but a model using a craftmanship value is most promising. This weights in the users credits given the amouth of contributions at a specific revision

$q^{n}=q^{n-1}+{\frac {H^{n}(\mathrm {A} ^{n})}{H^{n}(I^{n})}}*c^{n}(K^{n})\,$

The constant $c^{n}(K^{n})$ is the one of the

---

There are at least two different models possible for the article quality. One uses Shannons entropy for the article and the other uses the users credits to estimate a value, or more of a craftmanship value. Both can be used together and the model will adjust Shannons entrophy with the users credits.

Shannons entropy for any article version is

$H^{n}(I)=-\sum _{i\in I}h(p_{i})=-\sum _{i\in I}p_{i}*\log(p_{i})\,$

For the revision $n$ the set $I$ is assumed to be the set of unique words.

This value can be calculated as a iterative serie but is just as simple to calculate from the last version.

If we assume

---

As the article quality is an accumulating series, which at any given point is described as

$Q^{n}=H^{n}(I)+\rho \sum _{i=1}^{n}C_{i}(K)$

The article quality is denoted $Q^{n}$ at the given iteration $n$, Shannons entropy over the set $I$ is $H^{n}(I)$ and the user creditabillity over the user set $K$ is $\sum _{i=1}^{n}C_{i}(K)$

Which can be somewhat simplyfied and reformulated as discrete iterative series

$q^{n}=q^{n-1}+h^{n}+\rho c^{n}\,$

A quite mind-bogging conclusion is that the growth in quality is given by the additions of information until the users starts to brush up the article, at which point the article will slowly propagate to a mean given by the users credits.

The text is not proofread below this point

---

In the simple case $\rho$ could be a simple constant or it could be a more complex function

$q^{n}=h^{n}+\rho c^{n}+q^{n-1}\,$ Where $h^{n}$ and $c^{n}$ now is the differences to the previous version

The craftsmanship parameter $\rho$ describes how well the credits of the user transfers to the article. If this value is zero the article gains nothing from the user. If it is above zero the article will lock down faster than usual, if it is less than zero it will lock down slower than usual.

The penalty for reverted edits are in fact only easy to implement for direct trolling. An user adding edits which is reverted much later is more difficult. This can although be implemented by hashing all words in an article and assigning a count and the actual users. If all words are reverted then all users on that list gets a penalty. This works quite well for simple statements. This can be modeled as an approximation to the previous

$\sum c_{i+1}^{j}=\sum c_{i}^{j}-\beta ^{j}\,$

Missing description of the laug and how credits and fines works in this case. Overall d\there are credits and fines for single users, credits and fines for laug, and a craftsmanship parameter. A total of five parameters, or possibly, functions taking parameters. Should also include something about a priori knowledge (credits).