I know this isn’t the first time this has been discussed but I don’t think @MakerWorld weighed in on the previous threads and I would love to see the bugs worked out of the search algorithm.
Currently any popular model can add a popular keyword and rank at the top for that keyword, even if the keyword is completely irrelevant to the model. Here’s an example:
Currently it looks like the algorithm is only looking at two things, how popular the model is and if the word searched for is one of the keywords. I feel like it should weigh the title, description and keywords. I know those could all be manipulated but it will be more obvious someone is gaming the system when you see “Christmas Filament Rack System”.
That’s why I often find the model that fulfill my requests one page 3 somewhere between totally irrelevant models. That’s really not helpful.
All in all I think MW is focusing to much on models which already have a high amount of downloads. Why? Because new models are often iterations of older models which are just better. I got fooled by maybe the most popular dry boxes at MW - but the models I had before were quite better. They just have less downloads and therefore less rating in the search results.
But back to original topic: There should be no need for keywords at all. Google has eliminated the search tags in page head area a few years ago, as they found it, that this is justing for cheating and not helpful for the search index. They are mainly focusing on full text analysis of the web page and that is the way MW should go as well. Just remove the keywords in total.
I just tried it, 2 hours after your post, and the results are different. There is more to this than I thought! The search results may be personalized. Check this out:
ew, and makerworld forces you to have an account to download more than 1 model, so every user is about to be segregated into their own algorithm bubbles…
My “Christmas” search results are about 30% relevant in the first 50 or so models presented. The further down you scroll, the more relevant the results become.
They need to dial up relevance and dial down popularity by looking at click through rates per search term.
Ideally relevance would be counted by conversion rate, so if you look for “christmas” item and you click, download and print an item from search then that’s most likely item relevant to “christmas” search.
Another relevance trick would be to count collections, so if you look for “christmas” items and an item is in multiple “christmas” collections of various users, then it’s probably christmas-relevant.
this and “site:yeggi.com -cults3d” is usually my go-to.
im not saying people shouldn’t get paid for their designs but im not always going to shell out 1 to 40 dollars for something someone either stole or ripped the model from a video game claiming to have designed it themselves and isnt even something that’s actually printable.
i’m looking at you ffxiv “tomestone” (yes that’s how its spelled) “phone cases”.
I have another issue with MW search. I searced a model by the name of the file. It absolutely couldn’t find it for me. I guess the search funtion could use several im provements.
Playing Devil’s Advocate for a moment, as a retired IT Professional, I think this may be the result of two separate but related issues. On one side, it may be less to do with the algorithm used for searches, and more to do with what metadata is available to search against. Since MakerWorld doesn’t create the content associated with models that are posted, it’s up to the maker to provide as much keyword content as possible when formulating their posts. On the other hand, I do know if I do a search for a specific word I’ve used in the name of a model I’m posting, often the search function doesn’t manage to find it. And that is both odd and discouraging, as one then wonders how much content you’re not seeing when you perform a search on a given topic.
