Incorrect tags applied to package

[kichik]

What is the URL of the page with the issue?

https://constructs.dev/packages/@cloudsnorkel/cdk-github-runners/v/0.0.11?lang=typescript

What did you do?

I created a package with specific keywords and a readme that mentions others.

What did you expect to happen?

I expected only my keywords to be used as tags.

What actually happened?

My keywords were used as tags, but also keywords automatically found in README.md. Specifically it found Kubernetes because I mentioned another package that provides the same solution but for Kubernetes. My package doesn't use Kubernetes. It also found MySQL from the sample code even though the package doesn't have much to do with MySQL.

Web Browser

Chrome

OS

No response

Other information

No response

[Chriscbr]

So it turns out this bug is... sort of intended behavior at the moment. 😅 Automatically tagging packages based on the README is something we did to improve the visibility of packages when we launched Construct Hub, since many authors didn't add tags themselves. But as you point out, this has a lot of false positives. I think we can definitely look into tweaking these rules to be a bit less sensitive.

In the future these extra tags might become unnecessary once we have a way to search packages using a proper search index (e.g. OpenSearch).

[kichik]

Makes total sense. The categorization does help constructs hub. Maybe we can have a special tag or other way to prevent this? A special comment line in the readme?

[Chriscbr]

Yep - it's definitely possible to add some functionality that disables this behavior by adding a custom tag to the package, like construct-hub/do-not-autotag (and we'd probably want to hide this meta-tag on the front-end too). PRs are welcome.