Entity-Relationship Extraction Based On Graph Neural Networks

With the development of the Internet and knowledge graph technology,extracting entity relationships from network texts and knowledge graphs has become an important issue.Entity relationship extraction aims to extract various semantic relationships that exist between entities,such as the competitive relationship between business entities,the reaction relationship between compounds,and the relationship between network secu-rity entities.If a problem can be transformed into a problem of extracting relationships between entities,and the entity encoding is expressed,then the problem can be solved with the help of entity relationship extraction algorithms.In recent years,with the rapid development of graph neural networks,methods such as graph convolutional neural networks have achieved better results than traditional relationship extraction methods in many fields.Therefore,entity relationship extraction methods based on graph neural networks have also become A current research hotspot.This dissertation focuses on the extraction of entity relationships,based on graph neural networks,combined with specific domain problems,to carry out related entity relationship extraction algorithms.In general,this thesis mainly conducts research on the two tasks of extracting enterprise competition relations and chemical entity reac-tion relations.The main research work and contributions can be summarized into the following two aspects:(1)Based on the Internet Wikipedia data,the research on the extraction of corporate competition relations is carried out,and an extraction algorithm based on graph neural network is proposed.Previous entity relationship extraction methods only considered the feature information of the entity itself,such as the corporate text features in the Wikipedia corpus.There are also various hidden relationships between business enti-ties.These relationship features are conducive to predicting the competitive relationship between business entities,and there is no public data set for the prediction of corporate competitive relationship.In response to this problem,this paper firstly crawls a large amount of enterprise-related entity data from the Wikipedia website on the Internet,and constructs a corporate knowledge graph,and then proposes a method for predict-ing corporate competitive relations based on graph neural networks.This method uses graph neural network to extract the characteristics of enterprise entity relationship in the constructed enterprise knowledge graph,and realizes the extraction of enterprise com-petitive relationship.Finally,the two data sets are compared with the existing methods,and the experimental results prove the effectiveness of the method in this paper.(2)Aiming at the shortcomings of traditional methods in the extraction model of chemical entity reaction relations,this paper applies graph convolutional neural network to the task of extracting chemical reaction relations,and proposes a new compound reaction relation extraction method.The previous methods only paid attention to the relationship information between the compound entities,and ignored the sub-structure information within the compound and the interaction relationship between the entities.This paper constructs a compound knowledge map based on two public compound re-action relationship data sets,and then proposes a compound entity reaction relationship prediction model based on graph neural network.The model makes full use of the learning and representation capabilities of graph structure data features of graph neural network.We conducted experiments on two compound reaction relationship data sets,and the experimental results show that the method proposed in this paper has a higher accuracy rate than the existing methods.