Multiview Translation Learning for Knowledge Graph Embedding

Algorithm 2

Learning multiview graph embedding.

Input: A set of Knowledge Graph and Subgraphs , vector dimension k, global-view margin , global-view learning rate .

Output: A set of generated vector spaces .

(1)  ⟵ Initialize empty set for vector spaces

(2) fordo

(3)  ifthen

(4)    ⟵ global-view margin , global-view learning rate

(5)  else

(6)    ⟵ Randomly initialize local-view margin and learning rate

(7)  end if

(8)   ⟵ All entities and relations in respectively

(9)   ⟵ Initialize uniform (, ) for each ,

(10)  loop

(11)   //Sample a minibatch of size b

(12)    //Initialize the set of pairs of triplets

(13)   fordo

(14)   //Sample corrupted triplet

(15)   

(16)   end for

(17)   Update vectors w.r.t

(18)     end loop

(19)   Δ ⟵ ()//Trained parameters are saved as one vector space

(20)    ⟵  //Add vector space to sets

(21)     end for