基于BERT预训练模型的镁合金铸造缺陷命名实体识别

Named entity recognition of magnesium alloy casting defects based on BERT pre-training model

为提高镁合金铸造缺陷命名实体识别效果,构建铸造缺陷命名实体识别数据集,采用BERT预训练模型训练词向量,将训练后的词向量通过BiLSTM-CRF模型提取特征,识别镁合金铸造缺陷领域相关实体。结果表明,该模型相比其他语言模型,在自建的铸造缺陷语料数据集上的效果均有所提升,准确率、召回率和F 1值分别为94.81%、97.30%和96.04%。

This paper aims to improve the named entity recognition effect of magnesium alloy casting defect.The study consists of constructing the data set of casting defect named entity recognition;training the word vector by using the BERT pre-training model;extracting the feature of the trained word vector with BERT pre-training model to identify the related entities in the magnesium alloy casting defect field.The results show that compared with other language models,the effect of the model in the self-built casting defect corpus data set has been improved with the accuracy by 94.81%,recall by 97.30%and F 1 values by 96.04%respectively.