Unsupervised Learning · Recommenders

NMF Recommender

Factor the buys matrix into k hidden tastes — the reconstruction fills the gaps and recommends.

tastes k
V · who bought what Ŵ·H · reconstructed
iteration
0
error
state
press Factorise
bought (0–5)recommendation
tip: low k = broad taste groups; high k = memorises every buy
nmf.py