融合多特征图卷积神经网络的方面级情感分析  

Aspect-based Sentiment Analysis by Fusing Multi-feature Graph Convolutional Network

作　　者：郭荣荣 高建瓴 徐瑞涓 GUO Rongrong;GAO Jianling;XU Ruijuan(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵阳550025

出　　处：《小型微型计算机系统》2024年第5期1039-1045,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62166006)资助。

摘　　要：方面情感三元组抽取是基于方面级情感分析的一项新兴任务.针对目前大多数情感分析模型存在三元组抽取任务中忽略了单词之间的关系、序列模型的向量长度设定过短造成输入文本信息丢失的问题.本文提出了一种融合BERT和多特征提取的图卷积神经网络模型(MF-GCN).本文为三元组抽取任务定义了十种关系,然后使用双仿射注意力模块获取句子中每个词对的关系概率分布,并利用不同的语言特征关系形成不同的多特征图.最后使用图卷积神经网络在多特征图上重复图卷积操作以获得节点表示.MF-GCN在数据集上的F1值分别达到了60.47%、72.37%、62.09%、70.77%、61.37%、71.61%、64.51%、69.55%,表明本文模型在情感分析任务中具有优异的效果.Aspect sentiment triple extraction is an emerging task based on Aspect-based sentiment analysis.At present,most sentiment analysis models have the problem that the relationship between words is ignored in the triplet extraction task,and the vector length of the sequence model is too short to cause the loss of input text information.This paper proposes a graph convolutional network model(MF-GCN)that combines BERT and multi-feature extraction.This paper defines ten relationships for the triplet extraction task,then uses the double affine attention module to obtain the relationship probability distribution of each word pair in the sentence,and uses different language feature relationships to form different multi-feature maps.Finally,the graph convolution network is used to repeat the graph convolution operation on the multi-feature graph to obtain the node representation.The F1 values of MF-GCN on the dataset reached 60.47%,72.37%,62.09%,70.77%,61.37%,71.61%,64.51%and 69.55%,respectively,indicating that the model in this paper has excellent effect in sentiment analysis tasks.

关 键 词：情感分析 三元组抽取 多通道特征 图卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]