Improve search results accuracy

bjonglez · May 24, 2020, 4:52pm

Here is another request to improve the wiki

When searching for "release", I get a very long list of not-so-useful pages:

These pages are dummy pages like https://openwrt.org/releases/14.07?redirect=no (without redirect=no of course), and seem to match only because the path to them contains "release".

There are several problems here:

these dummy pages shouldn't appear at all in search results (the "release note" pages they redirect to already appear, but later in the list)
the search result should first show pages that actually contain "release" in their title. For instance, these ones.
also, results should probably appear with the most recent pages first

This plugin looks nice to configure search, but it does not really explain what you can configure exactly.

To hide pages I found this and this.

tmomas · May 24, 2020, 6:22pm

hidepages plugin: Is installed since long time. We use it to hide the dataentries from the search. Works also for the dummy pages you mention. I will add the hidepage tag to the dummy pages.
searchdefaults plugin: Not sure if this is what you expected: Options are limited, see https://github.com/cosmocode/dokuwiki-plugin-searchdefaults/blob/master/lang/en/settings.php

github.com

cosmocode/dokuwiki-plugin-searchdefaults/blob/ae23dc989c71637677ec481a1bcd23282abfa81e/lang/en/settings.php#L9-L16


$lang['default_sort'] = 'Default sort of the quick search';
$lang['default_sort_o_hits'] = 'Sort by number of hits, descending';
$lang['default_sort_o_mtime'] = 'Sort by last changed date, newest first';
$lang['default_time_limit'] = 'Default time limit';
$lang['default_time_limit_o_any'] = 'Any time';
$lang['default_time_limit_o_week'] = 'Past week';
$lang['default_time_limit_o_month'] = 'Past month';
$lang['default_time_limit_o_year'] = 'Past year';

On top of that: If I set the sort order to newest first, I don't see a difference to the normal sort order.

Sort by hits

Sort newest first

Plugin is installed, config see Admin -> Config -> searchdefaults.
Play around with it, try it out, uninstall it if it doesn't work for you as expected.

improving search experience by using the Search Tools

grafik430×987 19.8 KB

Narrow down your seach to a certain namespace, e.g. docs.
Note that you can also do this manually while typing in the quicksearch box, e.g. @docs to restrict your search to the docs namespace. See also https://www.dokuwiki.org/search
improving search experience by using dokuwiki pagenames with correct word separator
When the OpenWrt wiki was reorganized and merged wih the LEDE wiki, new pages have been created with incorrect word separators, e.g. https://openwrt.org/docs/guide-developer/releases/cutting-minor-releases

Correct would be https://openwrt.org/docs/guide-developer/releases/cutting_minor_releases (that's the way the wiki creates new pagenames when you add links to a page).

Not sure how much this influences the search experience, but it should be mentioned.
Make pages in https://openwrt.org/docs/guide-developer/releases/ appear earlier in the quick search
I guess this will be difficult. IIRC, higher namespaces also count higher for the search, so

https://openwrt.org/releases/
beats
https://openwrt.org/docs/guide-developer/releases/

This is how the quick search looks when the pages you mentioned are moved to the /releases namespace. I guess that is what you want?

grafik462×1146 66.1 KB

tmomas · May 24, 2020, 6:39pm

Done, quicksearch looking like this now:

bjonglez · May 26, 2020, 10:26am

Thanks, it's already much better when hiding these dummy pages!

I never noticed this "Toggle Search Tools" before. I see it also allows to change the sort order (only for the given search). However, it's probably broken, because the results do not make sense when sorting by "last modified":

https://openwrt.org/start?do=search&sf=1&q=release&srt=mtime

bjonglez · May 26, 2020, 10:39am

It seems that this sort function was introduced in:

I haven't found anything else, or reports that the sort order is broken. I wonder if there's something special with our installation, or if nobody else is using this feature...

bjonglez · May 28, 2020, 6:32am

Now I get it... the sorting option only applies to "Fulltext results", not " Matching pagenames".

It's not well-designed because "Matching pagenames" appears first

CHKDSK88 · June 8, 2020, 9:53am

@tmomas

How to use wiki search now?

The example:
I try find this page something about LB-LINK: https://openwrt.org/toh/lb-link/lb-link_bl-w1200 or https://openwrt.org/toh/hwdata/lb-link/lb-link_bl-w1200 .

The result is:

tmomas · June 8, 2020, 4:05pm

It's normal that you don't find the dataentry via the normal wiki search. They are intentionally hidden, in order to avoid flooding the search results with dataentries. Dataentries are accessible via the ToH datatables.

Regarding the devicepage not showing up in the search: Strange issue... After a simple edit the page now shows up. It's unclear to me what the root cause for this issue is. Well, a good error comes back