| Representation learning of knowledge graphs(KGs)aims to embed elements of a KG including entities and relations into continuous vector spaces,so as to benefit a variety of downstream machine learning tasks while preserving the overall structure of the KG.However,most of the existing knowledge representation learning methods only focus on learning the representation of structured information located in triples,while ignoring the significance of semantic information for knowledge representation learning.This paper will first introduce two types of semantic information and propose corresponding methods of knowledge representation learning.Relation Hierarchical Structure(RHS),which is constructed by a generalization relationship named sub Relation Of between relations.We propose a novel method named TransRHS to incorporate the RHS into KG embeddings.More specifically,TransRHS encodes each relation as a vector together with a relation-specific sphere in the same space and employs the relative positions among the vectors and spheres to model RHS.Entity types indicate the categories of entities.In general,an entity may have multiple types,and the types of head entity and tail entity in a relation are specified.Therefore,a given entity should show distinct properties in different scenarios.Inspired by this,we propose a model named TransET to take advantage of this information.More specifically,each type is encoded into a projection matrix to model this information.TransRHS and TransET are evaluated on two typical tasks including link prediction and triple classification.Experimental results demonstrate that the performance of these two models is much better than other exsiting methods.Compared to the baselines,the TransRHS has an increase of 20% to 50% and the TransET has an increase of5% to 40% in multiple metrics.This fully shows that these two kinds of information are significant for knowledge representation learning,and these two models can fuse them into KG embeddings. |
PDF Full Text Request