Search results on hub.docker.com no longer relevant

[bcrickboom]

Hello Docker Team!

We use Docker (and Docker Hub) to share our open source work with the community for several years (BTW, thanks for your great work!).

I went back from holidays a couple of days ago and the results given for a research on hub.docker.com are no longer relevant.

Until a couple of weeks ago, a simple search with the single word “orthanc” was returning a list of Docker images which was more or less like that:

4 to 5 “biocontainers” images (like https://hub.docker.com/r/biocontainers/orthanc-postgresql), I don’t know nothing about these images, I assume that the “OSS sponsoring” make them first in the list, but users can see that there are no longer maintained and there are not a lot of pulls.
osimis/orthanc image which was the “official” Orthanc image till the beginning of 2024
orthancteam/orthanc image which is the new “official” Orthanc image
some other images, from us or from other users…

This was fine to us, I would have been happy to see these “biocontainers” images at the bottom of the list, but that’s not a true problem.

Today, the same simple research give this:

image
image787×1861 135 KB

As you can see, the first images in the list are some images related to Thanos and so on (and still the biocontainers ones), osimis/orthanc being at the 8th place and the “official” orthancteam/orthanc image at the rank 23!

So I’m wondering, is there a change in the algo behind the research button?
Or is it a bug somewhere?

Thanks in advance for your feedback!
Regards,

[rimelek]

This time I couldn't explain it either as I could not find the search term in the description or any data of some images in the search result.

We discussed it on the community forum originally: https://forums.docker.com/t/search-results-on-hub-docker-com-no-longer-relevant/142982

One thing I noticed which could be normal but worth mentioning is that the search result is different when we log out.

[actualben]

The main issue is that current search results prioritize trusted content at the top, even when it's irrelevant to the search query.

For instance, when you search Docker Hub for open-webui, a well-known open-source project without an official image on Docker Hub, you have to sift through approximately 85 pages of unrelated results before finding the first relevant entry. The relevant results are community-contributed images, including my own, which mirror the open-webui project.

This highlights a problem: when only "community" images exist for a project, they become extremely hard to find because the "trusted" status of unrelated projects currently outweighs the relevance of search terms. While there seem to be exceptions for some highly popular community images, in general, community images are being effectively hidden. I understand the security concerns with using random images from Docker Hub, but it would be beneficial to strike a better balance or provide an option to "enable" community images and revert to the previous search result order.

[bcrickboom]

Hello,

I don't know if there is another issue on the same topic but it seems that something has changed.

Now, when we search for 'orthanc', only results containing the 'orthanc' word ares returned :-)
That's really better!

I think there is still some room for improvement: indeed, as you can see on the following screenshot, 3 first results are sponsored results. But then, 3 following results are not maintained and not popular images... The first really relevant result comes at the 7th place:

So, shouldn't we take "last update date" and "number of pulls" in account in the ranking of the results?

I hope this helps to improve Docker hub :-)

Thanks!

[ivan-curkovic]

Hi @bcrickboom,

Thank you for your valuable feedback! We're actively working on improving the search algorithm, and insights like yours are incredibly helpful in guiding our efforts.
I will write here when we release new updates.

Ivan Curkovic (Docker Hub team)