基于预训练模型与神经网络的军事命名实体识别

Military named entity recognition based on pre-trained models and neural networks

军事命名实体识别是军事情报分析和作战信息服务的基础。为了解决军事文本中分词不准确、形式多样以及语料库缺乏的问题,文中提出了一种基于预训练模型与神经网络的军事命名实体识别方法。通过BERT预训练模型构建词向量,利用BiLSTM神经网络处理上下文信息的优势,同时加入注意力机制,并通过条件随机场进行解码,完成了军事命名实体识别任务。在自构建的军事文本语料库上的实验结果表明,该模型F1值为92.14%,优于其他的传统方法。

Military named entity recognition is the basis of military intelligence analysis and operational information services. In order to solve the problems of inaccurate and diverse forms of word separation in military texts and the lack of corpus,propose a military named entity recognition method based on pretraining model and neural network. The military named entity recognition task is accomplished by constructing word vectors through BERT pre-training model,using the advantage of BiLSTM neural network to process contextual information,while adding attention mechanism and decoding by Conditional Random Fields. Experimental results on a military text corpus of self-acquired pieces show that the model has an F1 value of 92.14%,which is better than other traditional methods.