| 摘 要: 命名实体识别是信息抽取和知识图谱构建等任务的关键技术环节。近年来，药品不良反应(Adverse Drug Reaction, ADR)实体识别受到学界的极大关注。为了系统梳理ADR实体识别的研究进展及趋势，首先总结了ADR实体识别时常见的问题；其次将ADR实体识别研究划分为基于规则和词典、基于传统机器学习、基于深度学习和基于迁移学习方法四类并进行分析和对比，得出目前基于迁移学习方法的ADR实体识别性能较优；接着给出了ADR实体识别涉及的数据集以及评价指标；最后对ADR实体识别未来研究方向进行了展望。 |
| 关键词: 自然语言处理；实体识别；深度学习；神经网络模型；药品不良反应 |
|
中图分类号: TP391.1
文献标识码: A
|
| 基金项目: 国家级大学生创新训练计划项目(SZDG2021040). |
|
| Survey of Research on Entity Recognition of Adverse Drug Reaction based on Machine Learning |
|
ZHONG Yule, MA Shiwen, LU Haojie, HAN Pu
|
(School of Management, Nanjing University of Posts and Telecommunications, Nanjing 210003, China)
1715272757@qq.com; 1535328066@qq.com; 1071879399@qq.com; hanpu@njupt.edu.cn
|
| Abstract: Named entity recognition is a key technical link in tasks such as information extraction and knowledge graph construction. In recent years, entity recognition in adverse drug reaction (ADR) has attracted great attention in academic circles. In order to systematically sort out the research progress and trend of entity recognition in ADR, this paper proposes to summarize the common problems in entity recognition of ADR. Then, the ADR entity identification study is divided into four methods respectively based on rules and dictionary, traditional machine learning, deep learning and transfer learning models. After analyzing and comparing the four methods, it is concluded that the current method based on transfer learning has better ADR entity recognition performance. In addition, the datasets and evaluation indexes involved in ADR entity recognition are given. Finally, the future research directions of entity recognition in ADR are prospected. |
| Keywords: natural language processing; entity recognition; deep learning; neural network model; adverse drug reaction |
