摘要
方面情感三元组提取旨在通过提取方面词、观点词及情感极性的三元组来为方面级情感分析提供解决方案。以往的研究存在无法处理句子中方面词和观点词之间的一对多和多对一的关系,以及在不同的子任务中存在错误传播的问题。本文提出一种基于多通道图卷积网络的情感分析方法,通过设计多通道图卷积网络来捕获单词的词性组合信息和结构信息。使用图卷积网络在异构图上重复图卷积操作得到句法依存模块的输入,同时使用双仿射注意力模块获取句子中单词间的关系概率分布。研究选取的数据集Res14、Lap14、Res15和Res16上的仿真实验表明,在F1值上,本文模型与现有的基线模型相比,取得了更好结果。
Aspect sentiment triple extraction aims to provide solutions for aspect-level sentiment analysis by extracting the triples including aspect words,opinion words and sentiment polarity.Previous studies have been unable to deal with one-to-many and many-to-one relationship between aspect words and opinion words in sentences,as well as the problem of error propagation in different subtasks.This paper proposes a sentiment analysis method based on multi-channel graph convolutional network,which captures the part-of-speech combination information and structural semantic information of words by designing multi-channel graph convolutional network.The graph convolution network is used to repeat the graph convolution operation on the heterogeneous graph to obtain the input of the syntactic dependency module,and the double affine attention module is used to obtain the relationship probability distribution between words in the sentence.Experiments on data sets Res14,Lap14,Res15 and Res16 show that the proposed model achieves more significant results on F1 values than the existing baseline model.
作者
郭荣荣
高建瓴
徐瑞涓
戚玲珑
GUO Rongrong;GAO Jianling;XU Ruijuan;QI Linglong(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)
出处
《智能计算机与应用》
2024年第5期36-43,共8页
Intelligent Computer and Applications
基金
国家自然科学基金(62166006)。
关键词
自然语言处理
情感分析
图卷积网络
三元组提取
natural language processing
sentiment analysis
graph convolutional network
triple extraction
