When in 2008 I made some observations on language versions, it struck me that in some cases the wikisyntax and the "meta article information" was more KB than the whole encyclopedic content of an article. For example, the wikicode of the article "Berlin" in Upper Sorabian consisted of more than 50 % characters for categories, interwiki links etc. This made me largely disregarding the cooncerning features of the Wikimedia statistics.
Kind regards

Am Dienstag, 6. August 2013 schrieb Aaron Halfaker :
I am removing all HTML tags and comments to include only those characters that are shown on the screen.  This will include the content of tables without including the markup contained within.  In other words, I stripped anything out of the HTML that looked like a tag (e.g. "<foo>" and "</bar>") or a comment ("<!-- [...] -->") but kept the in-between characters, whitespace and all.  

It seems much more reasonable to me that the difference is due to the fact that Fabian's dataset is limited to a very narrow range of bytes.  To check this hypothesis, I drew a new sample of pages with byte length between 5800 and 6000. 

The pearson correlation that I found for that sample is 0.06466406.  This corresponds nicely to the poor correlation that Fabian found.  

I've update the plot[1] to show the difference visually.  


1. http://commons.wikimedia.org/wiki/File:Bytes.content_length.scatter.correlation.enwiki.png

On Tue, Aug 6, 2013 at 6:04 AM, WereSpielChequers <werespielchequers@gmail.com> wrote:
Thanks both of you,

I suspect that you two are using very different rules to define "readable characters", and for Aaron to get a close correlation and Fabian not to get any correlation implies to me that Fabian is stripping out the things that are not linked to article size, and that Aaron may be leaving such things in.

For reasons that I'm going to pretend I don't understand, we have some articles with a lot of redundant spaces. Others with so few you'd be correct in thinking that certain editors have been making semiautomated edits to strip out those spaces. I suspect that Fabian's formulae ignores redundant spaces, and that Aaron's does not.

I picked on alt text because it is very patchy across the pedia, but usually consistent at article level. I.e if someone has written a whole paragraph of alt text for one picture they have probably done so for every picture in an article, and conversely many articles will have no alt text at all.

Similarly we have headings, and counterintuitively it is the subheadings that add most non display characters. So an article like Peasant's revolt will have 32 equals signs for its 8 headings, but 60 equal signs for its 10 subheadings. 92 bytes which I suspect one or both of you will have stripped out. The actual display text of course omits all 92 of those bytes, but repeats the content of those headings and subheadings in the contents section.

The size of sections varies enormously  from one article to another, and if there are three or fewer sections the contents section is not generated at all. I suspect that the average length of section headings also has quite a bit of variance as it is a stylistic choice. So I would expect that a "display bytes" count that simply stripped out the multiple equal signs would still be a pretty good correlation with article size, but a display bytes count that factored in the complication that headings and subheadings are displayed twice as they are repeated in the contents field, would have another factor drifting it away from a good correlation with raw byte count.

But probably the biggest variance will be over infoboxes, tables, picture  captions, hidden comments and the like. If you strip all of them out, including perhaps even the headings, captions and table contents, then you are going to get a very poor fit between article length and readable byte size. But I would be surprised if you could get Fabian's minimum display size of 95 bytes from 6,000 byte articles without having at least one article that consisted almost entirely of tables and which had been reduced to a sentence or two of narrative. So my suspicion is that Aaron's plot is at least including the displayed contents of tables et al whilst Fabian is only measuring the prose sections and completely stripping out anything in a table.

Both approaches of course have their merits, and there are even some editors who were recent edit warring to keep articles they cared about free from clutter by infoboxes and tables.



On 5 August 2013 21:16, Floeck, Fabian (AIFB) <fabian.floeck@kit.edu> wrote:

thanks for your feedback Jonathan and Aaron.

@Jonathan: You are rightfully pointing at some things that could have been done differently, as this was just an ad-hoc experiment.  What I did was getting the curl result of "http://en.wikipedia.org/w/api.php?action=parse&prop=text&pageid=X"  and running it through BeautifulSoup [1] in Python.
Regarding references: yes, all the markup was stripped away which you cannot see in form of readable characters as a human when you look at an article. Take as an example [2]: in the final output (which was the base for counting chars) what is left in characters of this reference is the readable "[1]" and " ^ William Goldenberg at the Internet Movie Database".
Regarding alt text: it was completely stripped out. This can arguably be done different, if you see it as "readable main article text" as well.
You are sure right that including these would lead to a higher correlation. Looking at samples from the output, the increase in correlation will however not be very big, but