基于层次注意力机制的远程监督关系抽取算法研究
首发时间:2020-01-29
摘要:远程监督机制由于其使用机器自动标注数据,能减少大量标注人力的优点,逐渐成为了知识图谱构建中关系抽取任务的主要手段。目前,如何能够较好的提取句子特征,为句子分类(关系抽取)提供良好的分类依据,成为了远程监督领域的一个研究课题。为了解决这个问题,本文采用了称为层次注意力机制的网络结构,该网络结构将注意力机制组织为层次结构,以更好地应对数据噪声,捕获句子的特征。本文使用注意力机制作为句子和句袋这两个层次特征的主要的编码器,构建了一个抗噪能力较强的远程监督机制的关系抽取器。实验表明,该模型在捕获句子特征、增强泛化能力方面超过了现有模型。
关键词: 深度学习\ 自然语言处理 \ 知识图谱 \ 关系抽取\ 注意力机制\ 远程监督
For information in English, please click here
Distantly Supervised Relation Extraction with Layered Attention Mechanism
Abstract:Distant Supervision has gradually become a main method in Knowladgegraph construction due to it reduced the manpower of annotating data because it uses machines to automatically annotate data. At present, how to better extract sentence features and provide good classification basis for sentence classification (relation extraction) has become a research topic in the field of distant supervision. To solve this problem, this paper uses a network structurecalled the hierarchical attention mechanism, which organizes the attention mechanism into a hierarchical structure to better deal with data noise and capture the characteristics of sentences. In this paper, attention mechanism is used as the main encoder of two levels of features: sentence levle and sentence bag level, and a relational extractor with strong anti-noise ability and distant supervision mechanism is constructed. Experiments show that the model outperforms existing models in capturing sentence features and enhancing generalization capabilities.
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于层次注意力机制的远程监督关系抽取算法研究
评论
全部评论0/1000