中医文献的句子级联合事件抽取  

Sentence-level Joint Event Extraction of Traditional Chinese Medical Literature

作　　者：高甦[1] 陶浒 蒋彦钊 贾麒 张德政[2] 谢永红[2] GAO Su;TAO Hu;JIANG Yanzhao;JIA Qi;ZHANG Dezheng;XIE Yonghong(Beijing Normal University Hospital,Beijing 100875,China;School of Computer and Communication Engineering,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]北京师范大学医院,北京100875 [2]北京科技大学计算机与通信工程学院,北京100083

出　　处：《情报工程》2021年第5期15-29,共15页Technology Intelligence Engineering

基　　金：国家重点研发计划云计算和大数据专项“大数据驱动的中医智能辅助诊断服务系统”(2017YFB1002300)。

摘　　要：[目的/意义]中医领域存在大量的文献,这些文献中含有大量中医诊疗的知识。但这些知识往往存在于非结构化文本中,通过信息抽取技术将其转化成结构化文本,不仅能够提高效率,还可以进一步推进中医智能辅助诊疗的发展。[方法/过程]本文使用了联合事件抽取模型,利用BERT对中医文献进行字向量表示,并在此基础上加入Self-Attention层,最后与CRF模型结合,实现了对中医文献的句子级事件的初步抽取。[结果/结论]通过实验与Pipeline模型进行对比,结果显示,本文使用的模型F1值较Pipeline模型提高了14.2%。[Objective/Significance]There are many literatures in the field of Traditional Chinese Medicine(TCM),which contain massive knowledge of TCM diagnosis and treatment.However,the knowledge often exists in unstructured text.Transforming it into structured text through information extraction technology can not only improve efficiency,but also further promote the development of intelligent diagnosis and treatment with TCM.[Methods/Process]We use the joint event extraction model,representing the word vector of the TCM literatures with BERT.We add a Self-Attention layer on this basis,and finally combine with the CRF model to achieve the sentence-level events extraction from the TCM literature.[Results/Conclusions]The experiment is compared with the Pipeline model,and the results show that the F1 value of our model is 14.2%higher than that of the Pipeline model.

关 键 词：事件抽取 信息抽取 中医文献 联合事件抽取模型 

分 类 号：G35[文化科学—情报学]