Do I need Negative Sampling in Poisson Factorization?
I have seen some examples of using Poisson Factorization in rating prediction recommendation tasks.
But my data is a user(M)*item(N) boolean matrix
in which 1–> indicate user interact with item and 0–> otherwise.
How to correctly use Poisson Factorization in this?
- model only the ‘1’ in the matrix
or - model the ‘1’ along with Negative Sampling