The final representation is the form of combined layer embeddings.euβ=βk=0KβΞ±kβeu(k)β;eiβ=βk=0KβΞ±kβei(k)β
The model prediction is defined as the inner product of user and item final representations: y^βuiβ=euTβeiβ . It implies the similarity between the user and item.
Matrix Form:
A=[0RRT0β],E(k+1)=(Dβ1/2ADβ1/2)E(k)
R is a MΓN user-item interaction matrix. Each entries 1 if u is connected to i
Dis a (M+N)Γ(M+N) diagonal matrix, in which each entry Diiβ denotes the number of nonzero entries in the ithβrow vector of A
E is a (M+N)ΓTmatrix where T is the embedding size.
We easily make this as a code using torch_geomtric.utils (reference)