Research On Knowledge Graph Completion Based On Deep Learning

Knowledge Graphs is a kind of graph model.It describes the relationship between knowledge and modeling each item in the world.The problem of insufficiency is more and more important.Despite there is large number of structured entities,relationships and the facts about the real world or abstract concepts,knowledge graph is still of sparseness.To complement the knowledge graph,there are two effective ways.One way is that using natural language processing model to extract knowledge from the massive Internet data.The other is that reasoning and merging based on the information stored in the knowledge graph itself.With a brief introduction of the development of knowledge graph and its application,this paper introduces the current situation and classical results of knowledge representation learning algorithms used in knowledge graph completion.Moreover,a novel knowledge representation learning algorithm is proposed.The relationship-path containing rich semantic information and the text description information of the entity are integrated into the process of representation learning.Relationship-path is an effective enhancement to relationship in the knowledge graph,while text description information can increase the semantic information in the embedded representation of the entity.At the same time,we also explores different ways of link prediction work which is a typical process of knowledge graph completion.Using convolutional neural networks to capture more complex interactions between entities and relationships,including linear interaction,we can make more effective judgment on the reliability of the facts complementation.We implemented the new model on the classical data set.And the experimental results show that compared to the baseline models,our model achieved significant improvement.It shows that the relationship path and text description information can promote the embedding of knowledge graph.Convolutional neural networks can capture complex interactions between entities and relationships in a triplet.