Tickets :: [#8572] Learning (adaptive) autocomplete

6.0.0-alpha10

5/14/25

Summary	Learning (adaptive) autocomplete
Queue	IMP
Queue Version	Git master
Type	Enhancement
State	Accepted
Priority	1. Low
Owners
Requester	chuck (at) horde (dot) org
Created	09/12/2009 (5723 days ago)
Due
Updated	03/05/2011 (5184 days ago)
Assigned
Resolved
Milestone
Patch	No

03/05/2011 04:16:56 AM	Chuck Hagenbuch	Comment #5 Taken from Michael Slusarz State ⇒ Accepted	Reply to this comment
k, so I think we keep this around for now. Thanks!

03/04/2011 06:23:28 PM	Michael Slusarz	Comment #4	Reply to this comment
What do the two weighting methods you described actually do? They provide more intelligent search results based on a partial search string. But they don't learn/adapt over time based on user input, which I believe is the intent of this request.

02/20/2011 01:51:57 AM	Chuck Hagenbuch	Comment #3 Assigned to Michael Slusarz	Reply to this comment
What do the two weighting methods you described actually do?

12/10/2009 12:08:30 AM	Michael Slusarz	Comment #2 State ⇒ Feedback	Reply to this comment
Sort of have this in IMP - we now weight autocomplete by either liquidmetal.js (javacript search) or the PHP levenshtein function (server-side search). Doesn't account for learning, but provides more intelligent result ordering.

09/12/2009 03:51:09 PM	Chuck Hagenbuch	Comment #1 Priority ⇒ 1. Low Type ⇒ Enhancement Summary ⇒ Learning (adaptive) autocomplete Queue ⇒ IMP Milestone ⇒ Patch ⇒ No State ⇒ Accepted	Reply to this comment
This could potentially be useful anywhere we do autocomplete, depending on how it's implemented, but IMP address completion is the most obvious case. Implement a learning (or really just weighting) algorithm so that the user's auto-complete choices are used over a straight match. So if there are two options when the user starts typing "Dan", and the user always picks the 2nd one, we should show the 2nd one as the default option.