基于依存关系和双通道卷积神经网络关系抽取方法  

RELATIONSHIP EXTRACTION METHOD BASED ON DEPENDENCY RELATION AND TWO-CHANNEL CONVOLUTIONAL NEURAL NETWORK

作　　者：吴佳昌 吴观茂[1] Wu Jiachang;Wu Guanmao(School of Computer Engineering and Technology, Anhui University of Science and Technology , Huainan 232001, Anhui, China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《计算机应用与软件》2019年第4期241-246,267,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61471004)

摘　　要：关系抽取是自然语言中的一项重要任务,其结果对后续的信息抽取和自动问答系统有重要的影响。随着深度学习的日益火热,基于卷积神经网络的实体关系抽取已取得了不错的结果。不过词向量表示比较单一,提取的特征也有限。针对这个问题,将Word2vec训练的词向量和由自然语言处理工具得出的依存关系对分别作为模型两通道的输入向量,使用双通道卷积神经网络提取特征来实现实体关系抽取。该模型可以提取更深层的语义信息,并取得了比传统词向量更好的效果。Relation extraction is an important task in natural language, and its result has important influence on subsequent information extraction and automatic question-answering system. With the increasing popularity of deep learning, the entity relation extraction based on convolutional neural network has achieved good results. However, the word vector representation is relatively single, and the extracted features are limited. In view of this problem, the word vector trained by Word2vec and the dependency relationship obtained by the natural language processing tool were used as the input vectors of the two channels in the model. Two-channel convolutional neural network was adopted to extract features to realize entity relationship extraction. This model can extract deeper semantic information and achieve better results than traditional word vectors.

关 键 词：关系抽取 依存关系 卷积神经网络 双通道 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]