404261 - Autocomplete favours impractical items with the first one to two letters are typed

Status: RESOLVED FIXED

(Firefox :: Address Bar, defect)

Description

[Kroc Camen]

User-Agent:       Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9b1) Gecko/2007110903 Firefox/3.0b1
Build Identifier: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9b1) Gecko/2007110903 Firefox/3.0b1

When typing in the Location Bar, the new bookmarks/grep/history autocomplete favours impractical URLs when only one or two letters are typed. For example, if I type "O", for OSNews.com, a site I visit often, the first item on the list is "kroc.deviantart.com". This is not as efficient as FF2.x until I have typed two letters - os (However depending on URLs, the new autocomplete may not "home-in" until two or more letters have been typed)

Reproducible: Always

Steps to Reproduce:
1. Using your existing history, type one letter in the location bar
2. Notice that the first item in the list does not start with chosen letter and barely related at all
3. There is no step three.


Expected Results:  
For one (and potentially two) letters, Firefox should search by URL and frequency first and NOT grep the page title / bookmarks.

Comment 1

[Aru Sahni]

I agree with Kroc in that the improved location bar's heuristic needs some tweaking.  It should put a greater weight on the first letter of a url, while at the same time taking page title into account.

Also, can the www be ignored?  Whenever I type a "w" in the location bar, I want to see sites whose domain starts with a "w" (e.g. "worsethanfailure.com"), not sites I visit more often that have non "w" domains (e.g. "www.digg.com").

I like the idea of searching by title, but I (and many other, I hope) have been using the URL method for so long that any transition towards a title search needs to be conservative...

Comment 2

[(not reading, please use [e-mail] instead)]

confirming.

the plan is that together, bug #394038 and bug #395739 will help address this.

Comment 3

[Aaron Nowack]

Isn't this effectively bug #393678?

Comment 5

[Nickolay_Ponomarev]

> the plan is that together, bug #394038 and bug #395739 will help address this.

I don't think so, since the global frecency algorithm can still prefer an item that doesn't match the typed text in a sensible way (e.g. "t" in "http"), and adaptive learning is good, but doesn't work until the user provides enough feedback.

Comment 6

[Ed Lee :Mardak]

I've been using the adaptive stuff for a while and because it checks previous inputs first, when I type "p", the results I get are "Planet mozilla", "caPital one", "sqlite exPression", "javascriPt shell", "comPiler construction", "aPple".

Those results are more meaningful than ones that just match httP in all the urls, but then again, it's subjective if those results are more useful to a user looking for "p<something>".

Comment 7

[Aaron Leventhal]

I don't think the adaptive learning is as important as weighing higher matches with the the beginning of a URL/title and then next the beginnings of words. The adaptive learning would be frosting on top of that.

Comment 8

[Ed Lee :Mardak]

Now that both bug 393678 and bug 395739 are fixed, typing one or two letters should produce better results.