Page MenuHomePhabricator

Add link engineering: Allow filtering of links based on page attributes
Closed, DeclinedPublic

Description

We already support using categories, templates, page size and other characteristics as factors when choosing which articles to send to the link recommendation service (T261408).

This task is about processing links as returned from the service. For example, if the service returns a set of 10 links, we may want to remove some of those links from the recommendation set based on categories/templates/other characteristics which belong to the pages being identified as candidate links.

As @RHo noted, one use case is to exclude disambiguation pages as link targets

We could either implement this in the link recommendation service and pass the configuration from to-be-defined keys in NewcomerTasks.json, or we could do the filtering in refreshLinkRecommendations.php

Event Timeline

@MMiller_WMF could you let us know if you want to do this for the initial release?

@kostajh -- I remember we talked about how disambiguation pages were very unlikely to be recommended because the algorithm would notice that they are rarely links. I'm having trouble thinking of other things we would want to filter out. What other ones might there be?

@kostajh -- I remember we talked about how disambiguation pages were very unlikely to be recommended because the algorithm would notice that they are rarely links. I'm having trouble thinking of other things we would want to filter out. What other ones might there be?

Nothing I can think of either right now, so I'll decline this and we could reopen if we decide we want to pursue it later.