基于双句法交互图注意力网络的方面级情感分析
Aspect-level sentiment analysis based on dual syntactic interactive graph attention networks
杨长春 1刘昊 1张毅 1李艺1
作者信息
- 1. 常州大学计算机与人工智能学院阿里云大数据学院软件学院,江苏常州 213164
- 折叠
摘要
为减少利用未处理的短语树引入的关于方面词错误的句法信息,提出一种双句法交互图注意力网络模型.在现有短语树的基础上通过特定的句法拆分获得面向方面的短语子树,在此基础上,在短语树与依赖树之间利用各自的句法特点建立句法信息的交互通道,有效结合短语树与依赖树两棵句法树产生的句法信息.在3个公共数据集上的充分实验结果表明,双句法交互图注意力网络模型均优于当前的主流方法,验证了模型的有效性.
Abstract
To reduce the wrong syntactic information about aspect word introduced through unprocessed constituent tree,a dual syntactic interactive graph attention network(DSGAT)model was proposed.The aspect-oriented constit-uent tree was obtained by specific syntactic splitting on the basis of the existing phrase tree,and on this basis,the interaction channel of syntactic infor-mation was established between the constituent tree and the dependency tree by using their respective syntactic features,effec-tively combining the syntactic information generated by the two syntactic trees of the constituent tree and the dependency tree.Adequate experimental results on three public datasets show that the DSGAT model outperforms the current mainstream methods and the effectiveness of the model is verified.
关键词
方面级情感分析/图注意力网络/短语树/依赖树/句法信息/句法拆分/句法交互Key words
aspect-based sentiment analysis/graph attention neural network/constituent tree/dependency tree/syntactic infor-mation/syntactic splitting/syntactic interactive引用本文复制引用
基金项目
江苏省产学研合作基金项目(BY20221171)
江苏省研究生科研创新计划基金项目(KYCX22_3064)
出版年
2024
NETL
NSTL
万方数据