It depends on what you mean by user-to-user recommendations -- recommendations based on user-user similarity? No that is not how it works; it is based on matrix factorization. But that's just an implementation detail.
I think the question is rather, does it support the operations you want? If you want to recommend to new users, yes it can do that after just 1 data point for a user. Add the user-item pref via /preference
, then just call /recommend
. Or you can add all 100, then recommend. Or you can use /recommendToAnonymous
, yes.
If you mean computing most similar users -- yes and no. No there is no direct API method for this. You will have to run a second model where items and users are flipped, and then compute most similar items.