基于DeepMeSHⅡ模型的生物医学文献多标签分类  

MULTI LABEL CLASSIFICATION OF BIOMEDICAL TEXTS BASED ON DEEPMESH Ⅱ MODEL

作　　者：张子寒[1] Zhang Zihan(School of Computer Science,Fudan University,Shanghai 200433,China)

机构地区：[1]复旦大学计算机科学技术学院,上海200433

出　　处：《计算机应用与软件》2023年第10期130-135,149,共7页Computer Applications and Software

摘　　要：随着文本信息数据量迅速提升,多标签分类算法的开发研究逐渐成为热点。针对生物医学文献的文本内容,提出的DeepMeSHⅡ模型在目前最新进展DeepMeSH模型的基础上,将基于注意力机制的AttentionXML深度学习模型通过排序学习模型融合到该问题中,并利用期刊信息对模型进行特征改进。DeepMeSHⅡ模型应用效果很好,在生物医学文献语义索引国际比赛(BioSAQ2019)中取得了第一名的成绩。The rapid increase in the amount of text information data has prompted the development of multi-label classification algorithms.Aiming at the text context on the professional field of biomedicine,this paper proposes DeepMeSHⅡ model.Based on the latest development of the DeepMeSH model,it integrated the attention-based deep learning model AttentionXML and the state-of-the-art DeepMeSH model with the learning-to-rank framework.Journal information was used to improve model features.The application effect of DeepMeSHⅡ was good,and it achieved the first place in the international biological semantic indexing competition BioASQ2019.

关 键 词：多标签分类 语义索引 排序学习 注意力机制 信息挖掘 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]