Hi everyone.

I'm working on a web app to suggest users content based on their likes and dislikes with machine learning. I don't really plan on making it public, but if it were to be published, how should rate limiting be handled ? Request would come from the same server, but for several distinct users.
And by the way, would that kind of service be allowed ?

This has been attempted before, with various degree of success.

The rate limits are described on the API wiki page.
https://e621.net/wiki_pages/2425
The hard limit is two requests per second, although you should make an effort not to make more than one request per second over a sustained period.

The rate limiting is applied per IP address, so I would suggest that you set up some sort of queueing system for requests.
Perhaps, the database exports would be of some use to you? https://e621.net/db_export/
Although, those obviously do not include users' likes and dislikes.

Thanks for your answer ! I'll use the DB export, it'll work well with what I want to do.

Are there any rate-limiting for static files ? For now my application display pages of ~20 images, a new page is loaded when the user have seen all images, I think it should be fine, but better safe than sorry.