Research On Knowledge Extraction Technology Based On Deep Learning

Knowledge extraction is a critical technology in the field of natural language processing,which has strong research value and numerous application scenarios.In recent years,the rapid development of deep learning has brought knowledge extraction technology research to a new level.The current methods of knowledge extraction mainly use recurrent neural networks and convolutional neural networks,but these methods are not perfect.For example,recurrent neural networks lose part of the feature data in the process of modeling long-distance texts,and do not support parallel computing,which leads to training time.Greatly increase,the convolutional neural network will lose feature information in the pooling layer.This paper mainly focuses on the key technical issues in knowledge extraction.The main work and main results of the paper are as follows:(1)In the process of constructing the knowledge map of the ten ASEAN countries,traditional machine learning algorithms have low accuracy,low efficiency,and highly dependent feature design for Chinese natural language processing tasks,and propose a network structure based on BERT-Bi GRU-CRF To implement the named entity recognition(Named Entity Recognition,NER)method.Firstly,the BERT model is pre-trained to obtain low-dimensional vectors that can be understood by the computer,and then the word vectors are input into the bidirectional GRU-CRF neural network model to extract semantic feature information.In terms of data,it is proposed to use the Hash Cos similarity algorithm to more effectively dividing the data set,thereby reducing the impact of uneven data distribution.Experiments show that the BERT-Bi GRU-CRF model has achieved good results for the named entity recognition task in the ASEAN news text data set,and the F1 values of entities such as PER,ORG,and LOC have been improved a lot.(2)Relation extraction has an indispensable position in knowledge extraction.Aiming at the problems of slow training speed of the recurrent neural network and low accuracy of feature information,this paper combines the attention mechanism and Bi SRU model to construct a new type of neural network model for relation extraction.First,use the BERT model for word embedding training,extract the semantic representation of the word vector through the Bi SRU layer,and add the Attention mechanism to improve the accuracy of model relationship extraction.This model is used for comparative analysis with other models on the character relationship data set.Experiments show that this model is superior to other experiments in terms of F1 value and training speed,especially in the recognition of friends,cooperation,and relatives.The accuracy is higher,respectively.They were86.84%,84.67%,and 80.67%.