摘要
针对目前方面级情感分析存在句法依赖解析结果不准确、句法和语义信息没有充分利用的问题,提出一种双特征增强的图卷积网络。利用句法解析器中的依赖概率矩阵作为图卷积网络的邻接矩阵,减小解析结果的不准确性,对初始句法信息进行上下文动态加权增强提取句法信息的能力,对于语义信息,采用多头注意力机制构建动态语义图卷积网络,充分利用语义空间信息。实验结果表明,与基线模型相比模型取得了较明显的性能提升。
A dual feature enhanced graph convolutional network was proposed to address the current issue of inaccurate syntactic dependency parsing results and insufficient utilization of syntactic and semantic information in aspect level sentiment analysis.The dependency probability matrix in the syntactic parser was utilized as the adjacency matrix of the graph convolutional network,reducing the inaccuracy of analysis results.The initial syntactic information was dynamically weighted to enhance the ability in extracting syntactic information.For semantic information,a multi head attention mechanism was used to construct a dynamic semantic graph convolutional network,fully utilizing semantic spatial information.Experimental results show that compared with the baseline model,the model achieves performance improvement.
作者
夏敏捷
师钰博
樊银亭
XIA Min-jie;SHI Yu-bo;FAN Yin-ting(School of Computer Science,Zhongyuan University of Technology,Zhengzhou 451191,China;Central Plains Petersburg Aviation Institute,Zhongyuan University of Technology,Zhengzhou 451191,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2426-2433,共8页
Computer Engineering and Design
基金
国家自然科学基金-联合基金项目(U1504614)。
关键词
方面级情感分析
图卷积神经网络
多头注意力机制
概率矩阵
句法
语义
依赖树
aspect-level sentiment analysis
graph convolutional neural networks
multi-head attention mechanism
probability matrix
syntax
semantics
dependency tree
作者简介
夏敏捷(1974-),男,河南三门峡人,硕士,副教授,研究方向为自然语言处理;通讯作者:师钰博(1998-),男,河南周口人,硕士研究生,研究方向为自然语言处理情感分析;樊银亭(1975-),男,河南南阳人,博士,副教授,研究方向为人机交互技术、数据可视化与可视分析。E-mail:shiyubo@zut.edu.cn。
