Entity Relationship Extraction Based On Bi-LSTM And Attention Mechanism

With the advent of the Internet era for all,how to obtain the required knowledge quickly and accurately from large-scale,unstructured information has become a topic of widespread concern.In the process of mining and analyzing massive information,information extraction is a basic task.Information extraction is the process of extracting specific information from a piece of text and forming structured data.Entity relationship extraction is an important part of information extraction.It is mainly used to identify the semantic relationship between entities in the text.It is also the basis of many application systems,such as intelligent question answering and knowledge graph.Traditional relationship extraction methods are mostly based on rules or statistics,which is costly and not suitable for large-scale data processing.The extraction methods based on deep learning can automatically learn sentence features without complex feature engineering,and the extraction effect is better.But most current methods ignore the mining of text semantics.Therefore,based on the existing research foundation,considering that BiLSTM can capture the advantages of bidirectional semantic dependence and the attention mechanism can assign different weights to the semantic features of different functions,this paper combines the two to perform entity relationship extraction.In the feature extraction layer,four types of features,part-of-speech,entity recognition type,relative position and the context of entities are introduced.In order to obtain the main connection between entities,the shortest dependency path is also introduced.In order to make the model have the ability to judge the direction of the semantic relationship,the entity relationship direction presentation layer is introduced.Finally,the calculation result of different entity features are normalized and mapped to the probability of all semantic relationships through softmax,and the category of the entity relationship is output,so as to realize a complete relationship extraction model.The experiment uses the Sem Eval2010 Task8 data set,and sets up comparative experiments for the input features,model structure,counter-overfitting strategies,and the model itself.The results show that the F1 value of our proposed method is improved compared with the Bi-LSTM model that only combines the attention mechanism and the model after the introduction of the entity relationship direction presentation layer.