I host a community wiki related to geocaching. When looking at the visitor statistics, I notice that quite a lot of readers end up at the front page, never to be seen again. The front page is basically a list of links to various articles, and I have the feeling that its UX rating is very low.
I have som ideas for alternative layouts, but before jumping into a radical change, I'd like to do some split testing, to see if an alternative layout can direct the casual visitor in the direct the casual visitor in the direction of at least one other article.
I have a vague notion that such a test can be implemented with a scribunto macro, that alternates between different "Main Page" templates. The traffic volume is low enough that I have no caching in front of the wiki, so the split part should be straightforward. However, I'm a bit uncertain about how to capture the actual version presented to the user. Also, I'm a bit worried about how search engines will react to a page that changes randomly.
So I'm seeking input from others that have had similar considerations:
* How to implement the actual split. * How to record the version presented to the user. (I use piwik for analysis) * How to ensure that search engine crawlers see the same page for the duration of the split testing.
1) First, you probably want to split the contents of the main page into a set of templates, to make it modular. 2) Then, you can simply make multiple subpages of the main page each hosting a different selection and presentation of the same content. 3) Finally, to split users across the different versions, you can: a) use a constant title, say [[Main Page]], in which you {{alternate}} the transclusion of each subpage (version) daily or weekly (whatever you can easily tell apart in your stats tool); b) actually send users to the different titles with a random redirect e.g. via JavaScript or ("traditional" approach) some parser function magic either in [[MediaWiki:Mainpage]] or in a #REDIRECT on the [[Main Page]] (results rather unpredictable, depends on cache).
Nemo
On Jul 3, 2014 6:56 PM, "Anders Wegge Keller" wegge@wegge.dk wrote:
The traffic volume is low enough that I have no caching in front of the wiki, so the split part should be straightforward.
You still may have a parser cache.
The only existing infra for a/b testing I know of offhand is centralnotice. (not page body) Someone has probably also done so for UX. (CC Jared)
If you're changing only styles (not content) and don't care much about users without JS then try JS + eventlogging?
-Jeremy
mediawiki-l@lists.wikimedia.org