Research And Implementation Of Entity Relation Extraction Based On Deep Learning

With the development of the Internet and information technology,the amount of information on the network is increasing day by day,bringing great convenience to our lives.At present,there are many ways to record information,such as text,pictures,and videos,but we still mainly use text as the information recording method.Natural language processing is an interdisciplinary subject that integrates artificial intelligence and linguistics.By representing,analyzing,and processing textual information,it breaks through the interaction between computers and humans.Mining useful information from a large amount of text information has become the research direction of many researchers,and entity relationship extraction is a research direction in natural language processing.It predicts the relationship between named entities from sentences,transforms unstructured sentences into structured information,and then the computer further stores and calculates them.The traditional entity relationship extraction task mainly involves manually discovering and establishing features representing entity relationships,and performing relationship extraction through kernel functions.The results are affected by the selection of features and kernel functions,and can often only be applied to relation extraction in specific domains.With the application of deep learning in relation extraction,convolutional neural networks,recurrent neural networks,and attention models are introduced for relation extraction.Features that represent entity relationships in sentences are obtained through neural networks to complete the task of relation extraction.Compared with traditional algorithms,the entity relationship extraction algorithm based on deep learning can obtain the characteristics of entity relationships through the learning of neural networks,and has better results.In view of the above,this paper proposes a new entity relationship extraction model based on attention mechanism.This model segments sentences based on the entity.The attention is based on the relationship between the vocabulary and the entity in the sentence to obtain a vector representation of the sentence.It obtains the characteristics of the relationship between entities to realize the prediction of the relationship between entities in the sentence.Through the attention mechanism,the model can obtain the characteristic data related to the entity relationship,which can remove the noise data that is not related to the entity relationship in the sentence,and improve the model's ability to predict the relationship between the entities in the sentence.In order to further improve the model's results and speed up the convergence speed during training,based on the attention model in this paper,we borrowed the idea of implementing a generative adversarial model and realized the entity relationship extraction based on the generative adversarial idea.The model is divided into two parts: entity relationship extraction and generation of entity relationship features.During the training process,different loss functions are used for error back propagation.Based on this,the two parts of the model can be improved together during the training process,and a good entity relationship extraction model is obtained.Finally,this paper designs and implements a general entity relationship extraction framework.Based on it,it can quickly implement the entity relationship extraction algorithm and analyze and compare the results.The framework performs unified processing on the data set and analysis of results,and realizes the interruption and continuation of the model during the training process,and uses a modular design scheme.Based on this,we design and implement a model that predicts the probability of entity relations by inputting sentences containing entities,and then can directly observe the output of the model.