Research And Application Of Chinese Character Relationship Extraction Model Based On Deep Learning

Entity relationship extraction is an important downstream task in the field of natural language processing,in which the relationship extraction between person pairs is a typical task in entity relationship extraction.The identification and extraction of character relationships in the corpus can provide help and support for the construction of people’s social network,the generation of user portraits and the recommendation system.For the topic of Chinese corpus person relationship extraction,after analyzing the research background,current research status and related technologies of person relationship extraction,this paper completes the task of Chinese corpus person relationship extraction from label dataset preprocessing to deep learning network model,and finally establishes a prototype system of person relationship knowledge map,during which the following two main tasks are carried out:Firstly,this topic achieves good results on small-scale label datasets(about 30000 pieces of data)in the absence of labeling data from large-scale Chinese corpus.Based on the strong linguistic representation and feature extraction capabilities of the pretraining two-way language model BERT,a supervised pipeline relationship extraction model is proposed.In order to obtain the best model combination and illustrate the superiority of the model,this study designs three groups of comparative tests,Experiment1 compare the performance of Bi LSTM and Bi GRU in capturing bidirectional semantic dependence in the laboratory,and conclude that Bi GRU has better comprehensive performance;Experiment2 compares sentence-level attention mechanisms with different number of heads to improve model performance,and finds that the model with 12-heads of attention mechanism has a better overall effect;Experiment3 compares the performance of the best combination of models BERT-＿Bi GRU＿ATT(12)＿FC and other classical relationship extraction methods based on the previous two experiments。Experiments show that the proposed model is superior to other classical models in F1 values or other indicators,which strongly verifies the validity of the proposed extraction model in the Chinese corpus character relationship extraction task.Second,based on the relationship extraction model BERT＿Bi GRU＿ATT(12)＿-FC proposed in this study,set up a prototype system of Chinese person relationship knowledge map based on B/S architecture.In this paper,the author introduces the requirement analysis,system design,system implementation and system testing,and puts academic research into practical application.