Research On Entity Relation Extraction And Knowledge Graph Completion Based On Deep Learning

With the rapid development of the Internet and the explosive growth of network information,how to quickly and efficiently understand and utilize the information on the network has become the focus.Knowledge graph technology emerges in this environment.It extracts the data in the network into the form of fact triples,and can be widely applied in many fields such as artificial intelligence and information retrieval.In the construction of knowledge graphs,entity relation extraction and knowledge graph completion are two important sub-tasks.This paper studies the entity relation extraction and knowledge graph completion.Firstly,in order to solve the problem of introducing noisy data in the method of distant supervised extraction based on deep learning,this paper proposes an entity relation extraction model based on positive and negative instance interaction(PNIIM).The model inputs the dataset marked by the distant supervision method into the multi-rule joint data filter to divide into positive and negative instance sets,and iteratively trains the relation extractor with the positive and negative instance sets,and then re-divides the positive and negative sets in the process of continuous iteration.When the relation extraction model reaches the final stable state,the data of the negative set is eliminated as noise data,and the retained positive set is used as the training set to train a high-quality relation extractor.Secondly,many models in the existing knowledge graph completion methods only consider a single triplet information,and ignore the subgraph semantic structure information of the triplet in the knowledge graph.This paper proposes a knowledge graph completion model by combining subgraph convolution and tensor decomposition(SCTD).The model obtains the subgraph semantic structure information of the entity and processes it using the improved subgraph convolutional neural network,and then sets the system parameters to joint the subgraph convolution model with the tensor decomposition model to score the complement triplets.The triplet with the highest score is selected as a credible new triple to be added to theoriginal knowledge graph.Finally,the proposed PNIIM model and SCTD model are respectively compared and analyzed with other methods on different public datasets to verify it's effectiveness and superiority.