Page MenuHomePhabricator
Create Task
Maniphest T155559

Use implicit feedback to improve search rankings
Closed, ResolvedPublic
Actions

Description

This is an idea that came up at the Developer Summit, 2017

  • "Define" how we can use implicit user feedback to assess if user got the best result
  • Then how can we use that feedback to actually make the results more relevant?
    • OR rank those results in some manner

Things that might helps us get this AI built:

Related tasks:

Related Objects

Event Timeline

Halfak created this task.Jan 17 2017, 9:47 PM
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptJan 17 2017, 9:47 PM
Halfak updated the task description. (Show Details)Jan 17 2017, 9:55 PM
Halfak added a subscriber: EBernhardson.Jan 18 2017, 12:25 AM
debt triaged this task as Medium priority.Jan 19 2017, 11:06 PM
debt moved this task from needs triage to This Quarter on the Discovery-Search board.
EBernhardson added a comment.Jan 20 2017, 12:11 AM

I've also been testing implementations of clickmodels, specifically DBN.

Halfak edited projects, added artificial-intelligence; removed Machine-Learning-Team.Jan 20 2017, 8:38 PM
anuj_arora claimed this task.Nov 26 2017, 12:10 PM
anuj_arora added a project: Google-Code-in-2017.
EBernhardson added a comment.Nov 27 2017, 5:22 PM

This ticket is partially resolved, as the DBN mentioned previously is now in production for building search rankers.

That doesn't mean there is no other work to be done though, it is quite plausible there are better ways than what we are doing today to utilize impicit feedback for building MLR systems. A difficulty with doing this as part of google code-in will be that the raw data here is PII and thus requires an NDA to work with. Additionally google code in is rather short, but any full work on this from ideation through to online user testing will likely require a month or more of work.

Aklapper moved this task from Proposed tasks to Information needed on the Google-Code-in-2017 board.EditedNov 27 2017, 8:50 PM

@anuj_arora: Hi, thanks for adding the Google-Code-in-2017 tag. Do you plan to mentor this task, or what is your reason? :) If so, this task needs a better description.

Aklapper removed a project: Google-Code-in-2017.EditedDec 5 2017, 2:59 PM

@anuj_arora: Removing Google-Code-in-2017 as this looks out of scope. Please see https://www.mediawiki.org/wiki/Google_Code-in/2017

EBernhardson closed this task as Resolved.Sep 13 2018, 10:39 PM

This is generally done. There is a pipeline that attempts to extract user satisfaction from the click logs, and then runs an MLR pipeline that optimizes for ranking by expected satisfaction.

Content licensed under Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 unless otherwise noted; code licensed under GNU General Public License (GPL) 2.0 or later and other open source licenses. By using this site, you agree to the Terms of Use, Privacy Policy, and Code of Conduct. · Wikimedia Foundation · Privacy Policy · Code of Conduct · Terms of Use · Disclaimer · CC-BY-SA · GPL