Research On Entity Relationship Extraction Method Based On Graph Structure

The rapid development of the network has generated a large amount of unstructured text information,and it has become a huge and growing source of information.How to quickly and accurately get the knowledge that people want from the network information and standardize it into structured data has been widely concerned.The attention to the relationship extraction research has also brought new opportunities and challenges,and it has become an important forward direction for the academic community.The research in this direction is mainly to combine the machine learning algorithm and the deep learning algorithm to complete the task of extracting the relationship between entity pairs.Based on the current research,most of the relationship extraction methods have problems such as rough classification and many selected features.At the same time,the extraction result is not satisfactory and the time cost is high.Based on the fewer features,this paper transforms the relationship extraction task into two parts: trigger word extraction and relationship extraction,which achieves fruitful extraction results.The first part is the extraction of relational trigger words.Based on the slot filling algorithm,a graph-based relationship trigger word extraction method is proposed.The prepositional filtering module is added,and it determines whether the prepositions between the entity pairs within a certain distance are trigger words.The prepositional filtering module can reduce the overall running time.The part uses the dependency syntax analyzer to connect the sentence instances into a graph,and uses the Page Rank algorithm to calculate the importance score of each node in the graph.The optimal parameter values are obtained by the comparative experiment.The association rule mining algorithm calculates the confidence of the rules between each node and the entity pair.It is used to influence the comprehensive importance score.The comprehensive score calculation in the slot filling method is improved in the proposed method.The weighted sum of the Page Rank importance scores and the association rule confidence are the final comprehensive importance score.After the part of speech screening,all nodes with comprehensive importance scores are clustered by AP clustering algorithm to obtain the relationship trigger words.The second part is the classification of relationships.Its task is to transform the relationship extraction algorithm into a classification of the relationship trigger words.After introducing the basic classification algorithms such as SVM,feedforward neural network,RNN,LSTM,GRU,etc.,the best classifier is selected according to the experimental results using different classification algorithms.The input of the second module is the word vector converted by continuous words in a fixed window.Experiments show that the best performance is obtained when GRU is selected as the classifier for this relationship extraction.Experiments show that the best performance is obtained when GRU is selected as the classifier for this relationship extraction.Compared with the existing relation extraction methods,the method of this paper is simple and the number of features in the method is small.It can extract the relationship trigger words of entity pairs more accurately from sentences.Finally,this method is used to extract the relationship between people of the hand-labeled news corpus,so as to construct the character knowledge graph.The knowledge graph is stored in the Neo4 j database for visualization.The method was experimented in the Sem Eval2010 Task8 corpus and the F value reached 0.844.Compared with some traditional entity relationship extraction methods,the performance of this method is significantly improved.In conclusion,the research on the relationship extraction method of this paper has certain significance and reference value.