Recommending from a large catalogue

Since we can pre-compute V_u and V_m, why would we need to inference them in real time in order to recommend?
Couldn’t we just update V_u and V_m once in a while and use current values to recommend until the next update?

1 Like

Hey Laurent @ljb1706,

You must be asking about content-based recommendation, and yes, we can do it exactly as you said - store v_u and v_m for reuse, and update when needed.


1 Like