Research On Knowledge Representation Learning Based On Complex Relation Modeling

In recent years,with the rapid development of the Internet,the explosive growth of network data follows.Such large-scale information is not conducive to acquire and process.In order to facilitate the processing of such large-scale information and knowledge,and to facilitate the knowledge acquisition,knowledge Graph emerged as the times require.Knowledge Graph is a kind of knowledge bases that stores human knowledge in a structured way.Knowledge Graph defines a specific object or an abstract concept of the world as an entity in the Knowledge Graph,and the relationship between entities as a relation in the knowledge base.As a huge knowledge base,Knowledge Graph contains abundant information and knowledge,and can provide a wealth of prior knowledge for machines.Knowledge Graph has been applied in many fields,and plays a very important role in different fields.It has become an important resource to support the application of artificial intelligence,and has played a huge role in promoting the development of artificial intelligence.A typical Knowledge Graph can cover a large number of entity and relation facts.Although the scale of Knowledge Graph is very large and contains a large amount of structured knowledge,it is still not completed.And there is a problem of data sparseness.It is still an important task to constantly discover new facts,new knowledge,and expand Knowledge Graph.Machine learning methods are used to do the representation learning,which facilitates knowledge acquisition,integration and reasoning,and expand the Knowledge Graph.Such a research method has been recognized more and more in the industry,and has achieved remarkable results,and become a research hotspot.The construction of Knowledge Graph is helpful to improve the quality and performance of search,and has important research significance and value for information retrieval.In this paper,a translation-based multi-path embedding model(MPEM)is proposed.Three different algorithms of the model are proposed,named as the multi-step relationship path enhancement embedding model(MPEM),the embedded model MPEM(RC)for calculating the path reliability considering the relation constraint,and the path enhancement embedding model(TIC)considering the type information constraint for training.MPEM and MPEM(RC)use two different computing methods to calculate the reliability of the path.MPEM(TIC)takes the type information into account in the training to generate negative samples.Our model can not only achieve effective interaction between entities and relations when projection,capture their attributes and diversity,but also effectively utilize multi-step path information between entities to infer complex semantics implied between entities.Experiments on the standard data set show that our model is more effective than the previous models.