基于双向编码器表示模型和注意力机制的食品安全命名实体识别

针对于目前传统的命名实体识别模型在食品案件纠纷裁判文书领域的准确率不足的问题,在双向长短时记忆网络的基础上提出一种基于双向编码器表示模型(bidirectional encoder representations from transformers,Bert)和注意力机制的命名实体识别模型.模型通过Bert层进行字向量...

Food safety named entity recognition based on Bert and attention mechanism

Aiming at the problem of insufficient accuracy of the current traditional named entity recognition model in the field of food dispute dispute judgment documents, a named entity recognition model based on Bert and attention mechanism is proposed based on the two-way long-term and short-term memory network. The model pre-trains the word vectors through the Bert layer, generates word vectors based on the contextual semantics, the word vector sequence is input to the BiLSTM layer and the Attention layer to extract semantic features, and then the CRF layer predicts and outputs the optimal label sequence of the word, and finally obtains a food case dispute judgment The entity in the instrument. Experiments show that the accuracy rate and F1 value of the model in the legal documents of food disputes have reached 92.56% and 90.25%, respectively, and the accuracy rate has been improved by 6.76% compared with the BiLSTM-CRF model that is currently most used. The Bert-BiLSTM-Attention-CRF model can effectively overcome the ambiguity of missing words in the traditional named entity recognition model through pre-training of word vectors and fully integrate the context semantics, which improves the recognition of named entities in the field of food dispute dispute judgment documents. Accuracy.