[Epic] Build Media Matching API for bots/scripts
Open, Needs TriagePublic
Actions

Description

User story: As a bot writer on wikis, I want a reliable and robust API for matching highly relevant multimedia/images to specific articles, so that I can accurately add files (mostly images) to articles that my bots create or edit.

Acceptance Criteria:

Bots can call the Image Recommendations API to get a list of unillustrated articles on a given wiki and their image matches, both from the ImageMatching Algorithm and from MediaSearch
Bots can decide how many of the given matches to add to an article
Bots can call the Image Recommendations API to get a list of image matches for a given article from both the ImageMatching Algorithm and MediaSearch
The API returns a "confidence score" value in the API that bots can use to estimate how relevant/accurate the returned images are and decide which to add

During development, please test the following:

Use a bot on beta wiki to test the above features

Related Objects
Search...

Status	Assigned	Task
Open	nettrom_WMF	T261343 Dashboard of multimedia usage on the Wikipedias
Open	None	T267674 [Epic] Build Media Matching API for bots/scripts
Resolved	Cparle	T273062 [L] Create tool to manually test image recommendations POC results
Resolved	CBogen	T273092 Manually test image recommendations POC results
Resolved	Cparle	T281893 Repopulate testing tool after removing new "instance of" categories
Resolved	Cparle	T269852 [Epic] Interpret image search signal results
Resolved	Cparle	T271799 [L] Implement new search profile(s) based on image search signal results
Resolved	Cparle	T272710 Investigate whether the probability-of-an-image-being-good score is useful as a confidence score
Resolved	AikoChou	T274225 Multivariate logistic regression on search scores
Resolved	Cparle	T271801 Create mechanism for comparing search profiles using labelled data
Open	Cparle	T271803 [Epic] Improve mediasearch by using labelled data to create a model using elasticsearch learning-to-rank
Resolved	Cparle	T271806 Create elasticsearch featureset to get scores for each component in a mediasearch
Resolved	AnneT	T273882 [M] Estimate how many unillustrated articles on Cebuano and Arabic wikis would have matches in MediaSearch
Open	CBogen	T265771 Measure how multimedia content is added to Wikipedia articles
Resolved	matthiasmullie	T266067 [L] Create edit tags to measure multimedia edits to Wikipedia articles
Open	None	T282216 Add edit tags for each type of media for Hashtags tool
Open	nettrom_WMF	T275888 Measure success of Media Matching bots
Declined	None	T276135 Create a test bot to add images on a beta wiki

Event Timeline

• Ramsey-WMF created this task.Nov 10 2020, 7:06 PM

Restricted Application added a subscriber: Aklapper. · View Herald TranscriptNov 10 2020, 7:06 PM

• Ramsey-WMF added a project: SDAW-MediaSearch.Nov 10 2020, 7:24 PM

• Ramsey-WMF updated the task description. (Show Details)Nov 10 2020, 10:06 PM

Lsj added a subscriber: Lsj.Nov 12 2020, 10:08 AM

Cparle added a subscriber: Cparle.Nov 16 2020, 5:46 PM

@WDoranWMF @sdkim -- here's a task where the Structured Data team records what they want to surface to their bot writers for their use case. I wanted to point this out to ask whether it seems like this should somehow be part of the "Task API".

JKatzWMF added a project: Tech-Product API Roadmap.Dec 4 2020, 10:38 PM

CBogen edited projects, added Structured-Data-Backlog (Current Work); removed Structured-Data-Backlog.Dec 7 2020, 4:55 PM

CBogen added a project: Image-Recommendations.Dec 9 2020, 7:50 PM

CBogen moved this task from To Do to Epics on the Image-Recommendations board.Dec 9 2020, 10:08 PM

CBogen moved this task from Incoming to Epics on the Structured-Data-Backlog (Current Work) board.Jan 14 2021, 10:32 PM

CBogen moved this task from Backlog to MediaSearch-ImageRecs on the SDAW-MediaSearch board.Jan 27 2021, 2:12 PM

CBogen edited projects, added SDAW-MediaSearch (MediaSearch-ImageRecs); removed SDAW-MediaSearch.

CBogen added a subtask: T273062: [L] Create tool to manually test image recommendations POC results.

MMiller_WMF mentioned this in T254768: [EPIC] Image recommendations.Jan 28 2021, 8:16 PM

CBogen added a subtask: T273092: Manually test image recommendations POC results.Jan 28 2021, 9:03 PM

CBogen added a subtask: T272710: Investigate whether the probability-of-an-image-being-good score is useful as a confidence score.

CBogen added a subtask: T269852: [Epic] Interpret image search signal results.