摘要
药物不良反应检测对于用药安全与保证治疗效果有重要意义。传统通过检索人工构建的相关知识库来实现药物不良检测的方式低效且易错,很难带来真正的临床价值。随着人工智能技术近年来的快速发展,利用神经网络检测药物不良反应的方法展现出了巨大的应用潜力,但目前的研究普遍以先抽取实体,再判断实体之间关系的序列化方式解决问题,这样的方式会带来误差传递、信息冗余等问题。针对上述问题,提出了一种基于标注策略的多任务神经网络,将药物不良反应与药物相互作用两个任务建模为一个序列标注问题以提升模型的最终性能。实验结果表明,在相关国际比赛与公开任务的数据集上,提出的多任务神经网络在不同的评价指标上都取得了显著的提升。
The detection of adverse drug events is of great significance for drug safety and treatment. The traditional way, which detects adverse drug events by retrieving the artificial knowledge base, is inefficient and error-prone, so it is difficult to bring real clinical value. With the rapid development of artificial intelligence in recent years, the method of detecting adverse drug events using neural network has shown a practical potential. However, the existing research generally achieve the detection by extracting entities first and then classify the relation between entities in a sequential way. Such a solution will bring problems such as error propagation, information redundancy and so on. Aiming at solving these problems, a multi-task neural network based on tagging strategy is proposed which regards the adverse drug events and Drug Interactions two tasks as one sequence label task. The experimental results show that the proposed multi-task neural network has made significant improvements in data sets published from the related international competitions and open task.
作者
曹晓民
史瑞刚
CAO Xiao-min;SHI Rui-gang(Organization Department,Xi’an Aeronautical University,Xi'an 710000,China;Network and Information Center,Xi’an Aeronautical University,Xi'an 710000,China)
出处
《控制工程》
CSCD
北大核心
2020年第7期1151-1156,共6页
Control Engineering of China
