Feature Variable Relation Extraction Based On Deep Learning

With the continuous development of information technology,the global data volume is in the stage of continuous explosive growth.In this context,the importance of information extraction technology is becoming more and more prominent.Based on the analysis and research of subtask relation extraction of information extraction,this paper proposes a feature fusion relation extraction model(REFF)based on deep learning to solve the problem of insufficient extraction of incoming noise data and text features in remote supervision relation extraction.Model consists of three parts,respectively is based on the semantic features of BERT fusion,based on the grammatical features of GCN fusion,and part of speech feature fusion based on knowledge base.For a given package,a model of the three parts we got is a blend of semantics,grammar,and part of speech of the statement package distributed said,and then input it Softmax classifier resulting entity of the relationship between the tags.In this paper,the methods and research status of relationship extraction are described in detail,including the techniques of deep learning related to the model,the introduction of experimental data set,and the evaluation indexes.The main work of this paper is as follows:(1)The extraction of text features can greatly improve the effect of relation extraction.However,the existing methods of relation extraction only focus on one aspect of the features of the sentence,and do not integrate the semantic features of the sentence,so the grammatical features and part of speech features can be better represented.In order to obtain more abundant text features and improve the effect of relation extraction,we adopt the technology of deep learning to integrate the semantic features,grammatical features and part of speech features of text.(2)due to the automatic alignment of a large number of training corpus by remote supervision,a large amount of noise data has been introduced into the data,which has been another important factor restricting the effect of remote supervision method on relational extraction.Therefore,aiming at the problem of remote supervision relation extraction and the introduction of noise data,the paper adopts two Attention mechanisms to eliminate the influence of noise data on relation extraction.(3)in order to extract the characteristic variables of text better,we studied the BERT model,which has achieved great success in the field of natural language processing,and the graph neural network GCN,which has made an important breakthrough in the graph network data of non-euclidean space,and introduced additional edge alias information through the knowledge base in the relational extraction task.Finally,through the automatic extraction and fusion of text semantics,grammar and part of speech features,the effect of relation extraction is improved.Based on deep learning technology,this paper proposes the REFF model through the automatic extraction and fusion of text semantic features,grammatical features and part-of-speech features.The experiment removes the information of each part of the model by comparing the benchmark model and the comparison.The experimental results obtained are better than the comparison model,which shows the effectiveness of the REFF model that combines semantic,grammatical and part-of-speech features.