基于科学论文全文的论证结构关联数据构建及应用  

Argumentation Structure Linked Data Based on Full-Text Scientific Papers:Construction and Application

作　　者：曲佳彬 牛丽慧 欧石燕[3] QU Jiabin;NIU Lihui;OU Shiyan

机构地区：[1]滨州医学院卫生管理学院 [2]广东财经大学人文与传播学院 [3]南京大学信息管理学院

出　　处：《图书馆论坛》2025年第3期33-46,共14页Library Tribune

基　　金：国家社会科学基金青年项目“学术大数据背景下科学论文的论证结构抽取及应用研究”(项目编号:20CTQ009)研究成果。

摘　　要：为有效利用论文中的科学知识,解决学术界面临的信息过载和知识贫瘠问题,文章综合利用语义网技术和深度学习方法,设计一套面向科学论文全文的知识表示和抽取方法。首先从句子和实体两个层面构建面向科学论文内容的“论证结构-实体关系”本体;然后采用SciBERT-BiLSTM-Attention模型识别科学论文中包含的论证元素,并使用基于规则的方法识别科学陈述中的“实体-关系”以及其语境;最后以“阿尔茨海默症”领域论文为样本数据,构建“论证结构-实体关系”关联数据集,并通过语义检索、论证结构可视化和论证关系网络来展示知识服务模式。实验结果显示,通过上述方法能够有效表示和抽取论文中的知识,将论文中分散的科学知识转化为语义化和结构化的知识网络,不仅有利于促进科学知识交流和传播,还有助于智能化知识服务的开展。In order to effectively utilize the scientific knowledge contained in research papers and to address the problems of information overload and knowledge scarcity faced by the academic community,this article designs a knowledge representation and extraction framework for the full-text of scientific papers using semantic web technology and deep learning methods.First,an“argument structure-entity relationship”ontology for scientific papers is created from the perspective of sentence and entity.Then,a SciBERT-BiLSTM-Attention model is adopted to identify argumentation elements within scientific papers and the rule-based method is used to recognize entity relationships and their contexts.Finally,taking the papers in the field of Alzheimer’s disease as an example,the study develops an“argument structure-entity relationship”linked dataset,and demonstrates the knowledge service model through semantic retrieval,knowledge graphs,and argumentation relationship network.The experimental results show that the knowledge in the papers can be effectively represented and extracted by the above approach,and the dispersed scientific knowledge can be transformed into a semantic and structured knowledge network,which not only promotes the exchange and dissemination of scientific knowledge,but also contributes to the development of intelligent knowledge services.

关 键 词：论证结构 语义表示 SciBERT 关联数据 

分 类 号：G254[文化科学—图书馆学]