Page MenuHomePhabricator

Hide disambiguation pages when searching for the article subject in article guidance
Closed, ResolvedPublic2 Estimated Story Points

Description

As part of Article Guidance (T396029), assistance is provided to identify the subject for the article users are trying to create. The "article type selection" step (T415483) shows Wikidata items matching the proposed title for the user to select the relevant ones. Currently, these results include disambiguation pages, which do not represent a specific concept and it does not make sense to support them with an outline.

For example, searching for "Titanic film" results on two disambiguation pages showing: "Titanic" (Q738840) and "Titanic (film)" (Q28340274):

b24e11a4f1.catalyst.wmcloud.org_w_index.php_title=Special_NewArticle&mobileaction=toggle_view_mobile(Wiki Mobile) (5).png (320×568 px, 54 KB)

Interestingly, relevant search results like the specific Titanic films like Titanic (Q44578, "1997 film by James Cameron") appear in the results only when searching for just "Titanic" (but not when the search is for "titanic film").

Details

Event Timeline

Pginer-WMF triaged this task as High priority.
SBisson changed the task status from Open to In Progress.Mar 26 2026, 12:52 PM
SBisson claimed this task.
SBisson moved this task from Backlog to In-progress on the LPL Hypothesis board.

I'm adding a config for exclusion so we can add to it as we discover other similar cases.

What surprised me here is that the "Wikimedia disambiguation page" matches the "product" outline. It turns out it is matching via multiple paths but in essence, a web page is a piece of software which is a product. The product outline will probably be a remote match for a lot of concept.

I'm adding a config for exclusion so we can add to it as we discover other similar cases.

Perfect. Seems like a good and flexible solution.

What surprised me here is that the "Wikimedia disambiguation page" matches the "product" outline. It turns out it is matching via multiple paths but in essence, a web page is a piece of software which is a product. The product outline will probably be a remote match for a lot of concept.

Yes. I had similar impressions; I commented the following as part of the outline drafts:

Product (Q2424752) seems too general. As a result, searching in the test instance shows all kinds of results associated with the product outline (from books to disambiguation wiki pages).
Instead, we can consider some more specific types such as physical tool (Q39546), machine (Q11019) or electronic device (Q2858615)

Even if we readjust the scope of Product (or divide it into more specific outlines), I think the solution for redirects still makes sense and can still be useful.

Change #1261451 had a related patch set uploaded (by Sbisson; author: Sbisson):

[mediawiki/extensions/ArticleGuidance@master] Config to filter out article types

https://gerrit.wikimedia.org/r/1261451

Change #1261451 merged by jenkins-bot:

[mediawiki/extensions/ArticleGuidance@master] Config to filter out article types

https://gerrit.wikimedia.org/r/1261451