Knowledge Graph Attention Network for Recommendation

There are three main parts to this system: the embedding layer, which preserves the CKG’s structure and parameterizes each node as a vector; the attentive embedding propagation layer, which updates a node’s representation by recursively propagating embeddings from its neighbors and uses a knowledge-aware attention mechanism to learn the weight of each neighbor during a propagation; and the prediction layer, which combines the representations of a user and an item from all propagation layers, and outputs the predicted score.