结合领域先验词汇的远程监督关系抽取模型  

DISTANT SUPERVISION RELATION EXTRACTION MODEL COMBINED WITH DOMAIN PRIORI WORDS

作　　者：王会勇[1] 安康 张晓明[1] Wang Huiyong;An Kang;Zhang Xiaoming(School of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050000,Hebei,China)

机构地区：[1]河北科技大学信息科学与工程学院,河北石家庄050000

出　　处：《计算机应用与软件》2022年第8期34-43,50,共11页Computer Applications and Software

基　　金：河北省自然科学基金项目(F2018208116)。

摘　　要：在特定领域中,由于领域知识结构较复杂等领域因素,存在着缺少适用于特定领域的关系抽取模型以及缺少标注数据等问题。因此,提出一种基于先验词汇的分段池化卷积神经网络模型K-PCNN,该模型利用关系类别的领域先验词汇作为辅助特征,从而提高关系抽取性能。针对缺少标注数据的问题,提出基于远程监督的领域数据标注方法,利用领域三元组知识和领域语料文本进行数据标注。在标注的数据上进行实验,实验结果显示,提出的模型F1值高于基线模型,表明领域先验词汇知识的应用提高了模型的抽取性能,并能够适用于特定领域关系抽取任务。In specific areas,due to the complex structure of domain knowledge and other domain factors,there are some problems in domains,such as the lack of relation extraction model for specific domains and the lack of labeled data.Therefore,this paper proposes a piecewise pooling convolutional neural network model combined with domain priori words(K-PCNN).This model used domain priori words of relational categories as auxiliary features to improve relation extraction performance.For the problem of lack of labeled data,a domain data annotating method based on distant supervision was proposed,which could use domain triple knowledge and domain corpus to annotate data.Experiments were carried out on the labeled data.The experimental results show that the F1 of the proposed model is higher than that of the baseline model,which shows that the application of domain prior words knowledge improves the extraction performance of the model and can be suitable for domain specific relationship extraction tasks.

关 键 词：特定领域 关系抽取 先验词汇 远程监督 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]