Research On Entity Relation Extraction Based On Improved Bi-LSTM Model

Relation extraction is one of the key technologies to automatically construct knowledge atlas,and its fundamental goal is to extract semantic relationships between entities.The construction of knowledge graph provides a technical basis for intelligent information retrieval and semantic analysis,and has a very broad application prospect.Early entity relation extraction methods mainly rely on manual work and cost more resources.With the rising of deep learning,the method based on neural network model can better solve the problem of entity relation extraction.Neural network can automatically extract features without complex feature design engineering.There are still some problems in these entity relation extraction methods for unstructured text,such as the difficulty of accurately representing the context information,and the inability to fully utilize the entity relation features between sentences.This paper proposes a new relation extraction model based on the combination of neural network and attention mechanism.The basic network framework of this model is the combination of recurrent neural network and convolutional neural network.The relation extraction network model proposed in this paper is divided into two parts.The first part is to combine the Bidirectional Long Short-Term Memory Networks(Bi-LSTM)and Piecewise Convolutional Neural Network(PCNN)models that are good at dealing with long-distance dependency and add the attention mechanism.The second part is to insert Long Short-Term Memory Networks(LSTM)into the front end of Transformer model,and make full use of their respective advantages to combine,extract the features and word level features between adjacent words in the sentence corpus,so that the potential semantic information in the text corpus can be better excavated.At the same time,in order to enable the model to mine more feature information,the input layer also applies the entity to context feature information in addition to the word vector feature and location feature information,so as to better complete the entity relation extraction task.Finally,the model is tested in the dataset Sem Eval-2010 Task8 and Wiki80,and compared with several other classical models.Through experimental analysis,it shows that the model structure proposed in this paper has improved the performance of the results of entity relation extraction research,the experimental results reached 83.81% and 81.96%.