增强实体边界检测的医学命名实体识别

Medical Named Entity Recognition Based on Domain Knowledge and Position Encoding

针对中文电子病历报告中专业词汇较多导致的边界识别困难问题,文章提出了一种增强实体边界检测方法来更好地识别医学命名实体,即以实体边界预测为辅助任务,增强模型对实体边界的检测能力,提高模型性能。该文从两个方面增强了实体边界,一是通过在BERT与训练语言模型底层添加自制医学词典,增强模型对词汇边界信息的学习;二是以实体头尾预测作为辅助任务,进一步增强模型对实体边界的识别能力。在1个医学领域的公共数据集上进行了实验,相较于基线模型,F1值得到了1.96%的提升,说明该方法能有效检测实体边界,提升模型性能,验证了该模型的在医学领域的适用性。

Aiming at the difficulty of boundary identification caused by the large number of profssional vocabulary in Chinese clectronic medical record reports,this paper proposes a method to cnhancc cntity boundary detection for better identifying medical named entities.The method takes entity boundary detection as an auxiliary task,so that the model can enhance the ability of entity boundary recognition,and then improve the effect of entity recognition.This paper enhances cntity boundaries from two perspectives.One is to introduce a sclf-made medical dictionary to BERT for cnhancing the ability to learn boundary information;the other is to usc cntity head and tail prediction as an auxiliary task to further cnhance the models ability to identify entity boundaries.Experiments are conducted on a public data set in the medical field.Comparing with the baseline model,the Fl value is improved by 1.96%,indicating that this method can cffectively detect entity boundary,improve the model performance,and verify the applicability of the model in the medical field.