图神经网络系列二：NGCF & DiffNet

# 如图所示的用户free embeddinguser_embedding = self.user_embedding.weightfinal_item_embedding = self.item_embedding.weight# embedding的fusion融合层user_reduce_dim_vector_matrix = self.activation(self.user_fusion_layer(self.user_review_embedding.weight))item_reduce_dim_vector_matrix = self.activation(self.item_fusion_layer(self.item_review_embedding.weight))user_review_vector_matrix = self.convertDistribution(user_reduce_dim_vector_matrix)item_review_vector_matrix = self.convertDistribution(item_reduce_dim_vector_matrix)user_embedding = user_embedding + user_review_vector_matrixfinal_item_embedding = final_item_embedding + item_review_vector_matrix

# 用户的历史行为，user-item的交互user_embedding_from_consumed_items = self.bipartite_gcn_conv(x=(final_item_embedding, user_embedding), edge_index=self.edge_index.flip([0]), edge_weight=self.edge_weight, size=(self.n_items, self.n_users))# 用于存储融合社交的embeddingembeddings_list = [user_embedding]# 遍历N层，每层使用GCN对社交关系进行特征提取for layer_idx in range(self.n_layers):        user_embedding = self.bipartite_gcn_conv((user_embedding, user_embedding), self.net_edge_index.flip([0]), self.net_edge_weight, size=(self.n_users, self.n_users))        embeddings_list.append(user_embedding)# 将社交关系的提取的特征拼接起来，求和并且加上来自user-item交互提取的特征。final_user_embedding = torch.stack(embeddings_list, dim=1)final_user_embedding = torch.sum(final_user_embedding, dim=1) + user_embedding_from_consumed_items