摘要
针对方面级情感分析任务中,没有利用方面和意见词之间的间接依赖关系导致语法信息学习不完整,没有充分利用距离信息导致上文噪声词过滤不完全,对文本、语义和语法特征融合不充分的问题,提出了一种融合间接依赖和门控单元的双通道图卷积网络模型。该模型通过距离感知函数过滤上下文噪声,利用基于方面注意力机制的图卷积网络学习语义知识,使用融入间接依赖和距离信息的依存矩阵图卷积网络学习语法知识,通过双通道门控单元融合文本、语义和语法特征,将特征输入到线性层中得到情感极性。实验结果表明,该模型在两个公开基准数据集Lap14和Twitter上的准确率和F1值均有提升。
To address the limitations in aspect-level sentiment analysis,including incomplete syntactic information learning caused by the lack of utilization of indirect dependency relationships between aspect and opinion words,insufficient noise filtering in context caused by underutilized distance information,and inadequate fusion of textual,semantic,and syntactic features,a dual-channel graph convolutional network model that integrates indirect dependencies and gated units is proposed.This model filters contextual noise through a distance-sensing function.It learns semantic knowledge through a graph convolutional network based on the aspect attention mechanism and learns syntactic knowledge via a dependency matrix graph convolutional network that incorporates indirect dependency and distance information.The model fuses text,semantic,and syntactic features through a dual channel gated unit and inputs these features into a linear layer to determine sentiment polarity.The experimental results demonstrate that the model achieves improvements in both accuracy and F1-score on two publicly available benchmark datasets,Lap14 and Twitter.
作者
范瑞曌
唐非
FAN Rui-zhao;TANG Fei(School of Artificial Intelligence,Shenyang University of Technology,Shenyang 110000,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2388-2395,共8页
Computer Engineering and Design
基金
辽宁省教育厅科研项目面上基金项目(JYTMS20231224)。
关键词
方面级情感分析
图卷积网络
距离信息
方面注意力机制
语义信息
间接依赖
语法信息
双通道门控单元
aspect-level sentiment analysis
graph convolutional network
distance information
aspect attention mechanism
semantic information
indirect dependence
syntactic information
dual-channel gated unit
