Beta Features: Assistant Preview, Dragger AoL Improvements, EditableImage and EditableMesh, Face Capture, Gamepad Emulator, Import Queue, Improved Constraint Tool, Improved Mass Properties, Live Animation Creator, Multilayer Wrap Fix, New Studio Camera Controls, Next Gen Studio Preview, Preferred Text Size Setting, Scripts Are Non-Strict By Default, Texture Generator, UIDragDetectors, Updated Roblox Controls, Video Uploads
Reproduction Steps:
Open a blank baseplate with the toolbox window open.
Thanks for your report. This is actually behaving as expected and can happen as a result of our semantic matching where after it exhausts the closest matches some less relevant items will show up.
In the case of pure gibberish as you mentioned there are no relevant matches of course (unless you are really lucky) so these may show up.
I am going to close this ticket for now, but if it turns out these results are impacting a workflow that I may be missing please let me know.
The problem with that is the fact that these ‘less relevant items’ have in common 1 thing.
That they’re these 2 models very often.
Why should it be a few specific items that show up for less relevant searches? It should at least be somewhat as much as relevant as possible, even if that is a slim relevance. It doesn’t make sense.
Adding onto this post due to it’s anomalous nature at this point.
I’ve only seen these two specific models pop up no matter the search inquiry usually @ the end of any search inquiry list. For example if you search for anything that may be lesser known/not as widely used on the toolbox like, “HK416 HP Round” Which may I add does not have any results other than these 2 specific models
I wouldn’t add onto this post if it wasn’t a problem for other people but I’ve reviewed both models and over 3000+ people have dislike ‘raided’ the “Olympic Class (etc)” Model and over 2000+ on “Some Aura”.
I personally do not find this to be annoying or intrusive to my personal development work but I’d thought to look this up in the #bug-reports due to the sheer amount of people complaining about having this result on every query on the toolbox.
I do however find this to be unfavorable for the two creators of these items to have their toolbox creation/submission be dislike bombed due to simply just existing on the toolbox. I hope this provides broader scope and context to the bug.
Quick Edit: As a recommendation to ‘fix’ this issue for other users I think the toolbox should have a piece of text that says something along the lines of; “This doesn’t exist yet! Get creative and build it!” Or, “Search result provided no matches, relevant or misc objects displayed for inquiry.” Or just simply saying; “Nothing found!”
I hear the concern regarding these assets that commonly getting all the unwanted spotlight, we will discuss options internally on how to do this without impacting an assets ranking when it is relevant.
We are also currently testing a ranking model for which this side effect is less pronounced and more random which should help (though the primary reason we like this model is because it is better at relevance generally).
(To clarify, the model that I am referring to in that quote was the one in testing not the one you were seeing live in Store.)
Happy to share that the new semantic search model has now been released to Store. This model showed improvements in search relevance on semantic queries AND doesn’t seem to favor the Titanic or any assets in particular when searching random strings.
Thanks again for the report, if you see something like this again in the future please let us know!
I can confirm this update works.
I can tell with gibberish, it recognizes letters as words with them sometimes. Some of my gibberish with F in it will give me stuff like ferns, which is way better.
