Yea, I remember when QS's search was better than Spotlight. I think it
still can be with some tweaks to the ranker.
Below I have written up my proposed ranking logic. I think this could
compete or even beat current Spotlight results. I think this much more
closely matches how humans search and expect search results to be ordered.
No one runs a search expecting ranking based on this logic (see
underlines): ~/G*o*ogle D*r*ive/My
Stuff/www/*b*log/wp-content/upgrade/wordbook/wordbook/.
When we do a search we are usually interested in file name first, then
consecutive matches, beginning of word matches, then finally a pure
positional character ranking last to sort out the rest. The logic below is
basically doing this but written out more explicitly:
These are listed in priority/order of filtering:
*(using "orb" as an example search, X is wildcard character)*
1. Rank based on file name before path (or deepest folder name when
searching folders).
- ex: "/XXXXXX/orbis.app" should win against "/orbXXXXX/wordbook" for
'orb' because the file name is a better match. Path is just a tiebreaker.
2. Consecutive+beginning-of-a-word matches get priority over anything
else. (better than nonconsecutive with a better position as well as
consecutive that has better position but is nonbeginning).
- ex "XXX orbis.app"(consecutive+beginning) before
"XXorbis.app"(consectuive+nonbeginning) before
"oXrbis.app(nonconsective+beginning)".
3. A consecutive+nonbeginning of word beats a nonconsecutive with
better position.
- "Xorbis.app"(consecutive+nonbeginning) before
"oXrbis.app(nonconsecutive+beginning)"
4. A nonconsecutive with the user's first letter at the beginning of
at least one word beats non-consecutives that don't (essentially
textstarter's logic), and ones with beginnings of multiple words beats ones
with just one.
- "xxx oXrbis"(beginning) beats "XoXrbis"(nonbeginning with better
position)
- "oXXX rXXX bXXX" beats "xxx oXrbis"
5. For all other matches use current position ranking system plus a new
"closeness score" (see 6). Weighting of these two is a judgement call,
perhaps 70/30 is a good starting point.
6. Consider closeness of non-consecutive letters in addition to pure
position.
- ie Currently, XXXoXXXXXXXXrXXXXXXXXbX beats XXXXXXXXXXXorXbX
because of first o's position, but most would say the second is a better
match. Closeness could simply be total characters between letters.
7. Matches with same score but in a higher folder heirarchy position
beats ones in lower heirarchy
- when searching "o", /applications/Orbis.app beats
/applications/utilities/Orbis.app
On Monday, April 13, 2015 at 12:14:19 PM UTC-4, Etienne wrote:
>
>
> > Le 13 avr. 2015 à 17:51, random1destiny <[email protected] <javascript:>>
> a écrit :
> >
> >> On Monday, April 13, 2015 at 11:17:40 AM UTC-4, Etienne wrote:
> >>
> >> I'm not sure that filtering gobs of paths is a use case we support...
> >
> > I'm not sure what you mean by this. A decent logical search ranking is
> important to any application, let alone a launcher app whose sole purpose
> is getting to things quickly. It's the first thing anyone looks at when
> picking a launcher. In the post-Google world this use case should be
> assumed, especially when the application is built around searching for
> stuff.
>
> I meant that — though I see your point — QS main purpose is to find things
> fast, not really "searching" for stuff (at least that's how I understand
> it). I mean, don't use it to search for applications, just open them
> quickly using a few keys (and there was a time where QS surely beat the
> crap out of Spotlight), and the same for other kinds of objects, like
> Contacts — I don't really care about searching Contacts, I just care about
> having them at a few keystrokes' worth of my time. This is exemplified by
> our little motto "Act without doing, work without effort".
>
> So, maybe there's something I don't get about what you're actually doing
> ;-). The thing is, you'll *have* to "explain" to QS what you want to do,
> either forcefully through "Assign abbreviation...", or automagically
> through the Mnemonics system and repeated execution. As you can see the
> first time it will likely get things wrong because there's many things on
> you computer that could match "orb", and as it prefers matches closer to
> the beginning of what's matched because that's how it's done currently.
>
> Note that I do agree that the ranking looks fishy, so I'll try to take a
> look at what happens with file path... But I fear this requires some good
> thinking — a way of working around that mess would be some kind of "noise"
> filtering ("/Applications/" being kinda useless to match on if you get 1000
> items that match in there, as we can see, so that first 'o' should rank
> lower).
>
> Regards,
> Etienne Samson
> --
> [email protected] <javascript:>
>
>
--
You received this message because you are subscribed to the Google Groups
"Quicksilver" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/blacktree-quicksilver.
For more options, visit https://groups.google.com/d/optout.
