Research On Knowledge Graph Completion Algorithiom Based On Depth Learning

Since entering the Internet era,people's production and life have generated a large amount of data at all times.It has become a current research hotspot that how to quickly and efficiently understand and use this information.The emergence of knowledge graph technology solves this problem to some extent.In the knowledge graph,knowledge graph completion is an important task.This paper mainly studies the entity category completion technology and entity relationship completion technology in knowledge graph completion technology.In terms of entity category completion,this paper proposes an entity category completion model based on transfer learning to solve the problem which lack of labeled data.The model first constructs a mapping relationship model,and uses the mapping relationship model to calculate the semantic similarity between the entity category of the labeled data and the entity category of the unlabeled data.Therefore,a corresponding set of mapped labeled entity categories is constructed for each entity category of unlabeled data.Then,the combination of sentence vectors representing the mapping category set is input into the bidirectional long and short term memory network,and the unlabeled entity category is trained.The attention mechanism is constructed according to the semantic distance between each category in the mapping category set and the corresponding unlabeled category,and the entity classifier is used to classify the unlabeled entities,so as to achieve the purpose of completion.Experiments show that the proposed completion model solves the problem which entity category completion without labeling,and the completion effect under the condition of no labeling and sparse labeling is significantly improved.In terms of entity relationship completion,this paper proposes an entity relationship completion model which combining attention mechanism and convolutional neural network.The previous convolutional neural network model only used filters to extract text features for completion with out considering the problem which the different reference values of different feature information.To solve this problem,the model proposed in this paper firstly extracts the feature information of head entities and relations by using convolutional neural network,then further extracts the feature from two aspects of channel and space through attention mechanism,and assigns different features corresponding weights to complete the tail entity.Experiments show that the completion model proposed in this paper can effectively deal with the problem of entity relationship completion.