基于BERT-BiLSTM-CRF的突发公共卫生事件抽取研究

Event Extraction of Public Health Emergencies Based on BERT-BiLSTM-CRF

针对突发公共卫生事件的结构化抽取,首先定义了突发公共卫生事件的8种子类型及其结构,在此基础上提出基于BERT-BiLSTM-CRF的事件抽取模型.模型利用了BERT的多头自注意力网络和可微调特性,以管道方式连接事件分类、触发词识别和论元角色分类模型,完成了事件的结构化抽取.在基于新浪微博构建的语料集上的对比实验结果证明了该模型的可行性和先进性.模型的不足之处在于管道方式不可避免地会加大误差累积.

Aiming at the problem of structured extraction of public health emergencies,firstly,the 8 subtypes and structure of public health emergencies are defined.On this basis,an event extraction model based on BERT-BiLSTM-CRF is proposed.Used multi-head self-attention network and fine-tuning characteristics of BERT to connect event classification,trigger word recognition and argument role classification by pipeline,the structured extraction of events is completed.The comparison experiment results based on corpus set up by Sina micro-blog show that the model is feasible and advanced.The disadvantage of the model is that the pipeline mode will inevitably increase the error accumulation.