Quality Assessment Index

This topic relates to this very wikiHere be edit wars, censorship and template spam.

All pages by quality and importance

Quality	Importance
	High	Medium	Low	Unrated	total
Gemerald	32	10	6	0	48
Gem	67	86	79	3	235
Topaz	12	38	53	2	105
Ruby	8	23	32	0	63
Coal	3	20	36	6	65
Dust	0	0	28	0	28
Fossil	0	6	58	0	64
Total	122	183	292	11	608

This index is a combination of two separate ratings for every page that has a Quality header. Quality is the editor-given "rating". Importance describes how central the subject is to the wiki and is calculated automatically by this bot, meaning it cannot be tampered with by editors. Nothing in this page requires manual upkeep.

Importance is worked out once a month. The largest factors are how often a page is read (average monthly views over the last three complete months) and how many other pages link to it. Smaller factors are the number of editors and edits over the past year, and the word count of wiki prose. Each factor is compared against every other assessed page, and the combined result places the page in the top 20% (high importance), next 30% (medium importance) or the bottom 50% (low importance). Because ranking is relative, pages can move between classes when the rest of the wiki changes, even if their own numbers remain relatively static. Nu-pages with less than a full month of view data are unrated until enough info can come in for a reasonable placement.

This bot does not edit pages or talk pages to issue any categories. Instead, it writes a json dump to its own user space, which is read by a module invoked in each quality template to issue the correct rating category. The table above is a matrix of the index.

Formula

Nophono cares.

For every assessed page $p$, the bot collects five values:

$V_p$

average views per month over the last three complete months, counting only months since the page's first recorded view

$L_p$

number of distinct pages in the main and Namefag namespaces linking to $p$, redirects included

$D_p$

monthly editor counts over the last twelve complete months, summed

$E_p$

number of edits over the last twelve complete months

$W_p$

number of words of readable prose

Each value $x\in \{V,L,D,E,W\}$ is converted into a percentile among the $N$ scored pages:

$P_x(p)=\frac{100}{N}(|\{q:x_q<x_p\}|+\frac{1}{2}|\{q:x_q=x_p\}|)$

The score is the weighted sum of the five percentiles:

$S(p)=0.45P_V(p)+0.30P_L(p)+0.15P_D(p)+0.05P_E(p)+0.05P_W(p)$

pages are ranked by $S$ in descending order. With $r(p)$ the rank of $p$ starting at 0:

$\text{class}(p)=\{\begin{array}{cc}\text{High}& r(p)<\mathrm{round}(0.2N)\\ \text{Medium}& r(p)<\mathrm{round}(0.2N)+\mathrm{round}(0.3N)\\ \text{Low}& \text{otherwise}\end{array}$

Pages with no view data in the three-month window are unrated and not counted in $N$.