摘要
深度学习在自然语言处理方面取得了显著成效,为生物医学领域的信息抽取带来新的研究范式。本研究旨在系统调研生物医学语义关系抽取方法、分析其发展历程,为深度学习方法的进一步运用提供基础和启示。通过检索Pub Med、Web of Science和IEEE数据库,以及Bio Creative、Sem Eval等重要测评网站,遴选出具有代表性的抽取方法,并从目的、方法、数据集和效果四个维度进行分析。经过系统梳理,可将生物医学语义关系抽取方法分为三个阶段:基于知识、传统机器学习和深度学习。将先验知识和领域资源恰当地融入到深度学习模型中,是进一步提升语义关系抽取效果的探索方向。
Deep-learning has made remarkable achievements in natural language processing (NLP), and is bringing a new research paradigm to information extraction in biomedical field. This paper studies the extraction methods of biomedical semantic relations and analyzes its development progress and principles, which may serve as foundation for further application of deep learning. After retrieving relevant information from PubMed, Web of Science, IEEE, and other important websites such as BioCreative and SemEval, representative methods are selected and analyzed from four dimensions of purpose, approach, dataset and performance. Extraction methods of biomedical semantic relation can be divided into three stages: knowledge-based, traditional machine learning- based and deep learning-based. It is a new exploration effort to enhance the extraction effect of semantic relations by introducing prior knowledge and domain resources into deep learning model properly.
出处
《图书馆论坛》
CSSCI
北大核心
2017年第6期61-69,共9页
Library Tribune
关键词
语义关系抽取
生物医学
深度学习
卷积神经网络
自然语言处理
semantic relation extraction
biomedicine
deep learning
convolutional neural networks
natural language processing