Named Entity Recognition Based On Attention Mechanism In Dual-channel Neural Network

Named entity recognition plays an important role in natural language processing.Its purpose is to identify entities with specific meaning in text and classify them according to entity types.Aiming at the problem that statistical methods need a lot of time to construct features and deep learning methods lack rich input semantic information,this paper proposes a dual-channel neural network named entity recognition model based on attention mechanism.The commonly used model BiLSTM-CRF to solve the sequence labeling problem is used as a benchmark,and the benchmark model is improved to make it more suitable for the Chinese named entity recognition task.The main research contents are as follows:(1)A named entity recognition model based on dual-channel neural network is proposed to solve the shortage of semantic information input in named entity recognition(DW-BiGRU-CRF).Firstly,word2vec and GolVe tools are used to convert the experimental corpus into the corresponding vector representation as the input data of model channels 1 and 2.Then,BiGRU network coding is used to extract the semantic features of the vector.Finally,CRF is used to train and learn the relationship between the word vector and the output label to predict and output the best label sequence.Experiments show that the average accuracy,average recall and average F1 value of DW-BiGRU-CRF named entity recognition model in People's Daily corpus are 94.77%,93.51%and 94.13%,respectively.(2)Based on the DW-BiGRU-CRF model,attention mechanism is introduced to solve the influence of redundant information on named entity recognition,and a DW-BiGRU-ATT-CRF named entity recognition model is proposed.Attention mechanism is added to the neural network training to generate the weighted semantic representation of feature vectors,obtain more valuable contextual semantic features,and enhance the connection with the current entity-related contextual information,so as to improve the accuracy of entity recognition.Compared with DW-BiGRU-CRF model,the average accuracy,average recall and average F1 value of DW-BiGRU-ATT-CRF model were increased by 1.2%,0.75%and 0.98%,respectively.Experimental results show that the proposed DW-BiGRU-ATT-CRF named entity recognition model can accurately and effectively identify named entities in natural language.In this dissertation,the dual-channel neural network is innovatively used to enrich the semantic information of the model input,and the attention mechanism is introduced considering the different influences of each word in the natural statement containing the named entity.The attention mechanism is used to assign different weights to the semantic feature vector extracted by BiGRU,so as to improve the accuracy of model identification.Figure[20]Table[18]Reference[62]...