# How SVD factorisation -based recomendation algos deal with new user interaction

Classic SVD and SVD++ alogritms generate predictions based on a current known ratings only for known users and known items. But I need to make prediction for some new user on the old items. In the other words:

• I have the interactions matrix between N users and M items. And train the SVD model for them.
• I have a new N+1 user with known interactions on M items.
• I want to predict recommendations for the new user based on their interactions.

I know some ratings for this user and can make projection of his interactions into the latent features space. But it is only an intuition that can be used only for very clear and straitforward SVD approach. I can not find currently find any implementations and papers about this problem i have found just wiki https://en.wikipedia.org/wiki/Matrix_factorization_(recommender_systems) knows about this problem:

SVD++ has however some disadvantages, with the main drawback being that this method is not model-based. This means that if a new user is added, the algorithm is incapable of modeling it unless the whole model is retrained. Even though the system might have gathered some interactions for that new user, it's latent factors are not available and therefore no recommendations can be computed

And also i have seen in the implicit variant of SVD implementation that new users projection are calculated much more complicated way as just dot product between the user interactions and items.

Can anybody suggest some approaches for solwing this problem ?