Research And Application Of Relation Extraction Based On Graph Neural Network

With the rapid development of information technology,natural language processing technology is increasingly used in all aspects of life,such as the intelligent question and answer of Xiao Ai,Google's knowledge graph,Baidu's search engine,etc.These tasks use information extraction without exception,and information extraction is the basis for quality assurance in the subsequent steps.The relationship extraction used to extract the associated information between the entities in the sentence is an important subtask of information extraction.Therefore,its correctness is of great significance to the subsequent tasks.However,due to the complex sentence structure and numerous entities,how to improve the effect of relation extraction has always been the focus of scholars.At present,most of the researched relationship extraction tasks are still aimed at solving binary relationship extraction tasks,and most of them are based on sequence-based LSTM models.Considering that graph neural networks can better characterize the features of nodes in sentences,this thesis adopts entity-based associations Attention graph convolutional network relation extraction model solves binary and multivariate relation extraction tasks.In the research,it is found that entity information and location information have a great influence on the effect of relationship extraction.In order to further improve the accuracy of binary and multivariate relationship extraction,based on the graph convolutional network relationship extraction model based on entity association attention,this thesis proposes The multi-attention mechanism merges the relation extraction model to obtain a more complete vector expression.In addition,predicates often play an important role in relation extraction tasks,so dependency syntax information is often used to obtain structural dependencies between entities,but the information of a single dependency syntax tree is more limited.In order to solve this problem,and to further improve the effect of binary and multiple relation extraction tasks,a multi-dependency syntactic information fusion relation extraction model that integrates the characteristics of multi-dependency syntax trees is proposed.The three algorithms proposed in this thesis have conducted comparative experiments on the Semeval data set and the data sets of the electric power field and the medical field.The results show that the extraction model based on multi-dependency syntax information fusion relationship performs better than other algorithms.Finally,it is used in the subsequent construction and application process of power metering knowledge graph.