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

作　　者：杨长春[1] 刘昊 张毅 李艺 YANG Chang-chun;LIU Hao;ZHANG Yi;LI Yi(School of Computer Science and Artificial Intelligence,Aliyun School of Big Data,School of Software,Changzhou University,Changzhou 213164,China)

机构地区：[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

出　　处：《计算机工程与设计》2024年第8期2503-2512,共10页Computer Engineering and Design

基　　金：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)。

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。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 information was established between the constituent tree and the dependency tree by using their respective syntactic features,effectively 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 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]