| 王红霞,张佳慧,聂振凯.基于混合注意力和类型感知的方面级情感分析[J].高技术通讯(中文),2025,35(3):262~272 |
| 基于混合注意力和类型感知的方面级情感分析 |
| Aspect-based sentiment analysis based on hybrid attention and type-aware |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 03. 004 |
| 中文关键词: 方面级情感分析； 注意力机制； 方面感知注意力； 类型感知图； 图神经网络 |
| 英文关键词: aspect-based sentiment analysis (ABSA), attention mechanism, aspect-aware attention, type-aware graph, graph neural network |
| 基金项目: |
| 作者 | 单位 | | 王红霞 | （沈阳理工大学信息科学与工程学院沈阳 110159） | | 张佳慧 | | | 聂振凯 | |
|
| 摘要点击次数: 134 |
| 全文下载次数: 91 |
| 中文摘要: |
| 为解决方面级情感分析（aspect-based sentiment analysis，ABSA）任务中，未充分利用依赖树中的句法信息及语义信息提取不充分等问题，提出了基于混合注意力和类型感知的双图卷积网络模型。首先，设计了混合注意力模块，用于更全面地提取句子的语义信息，该模块采用方面感知注意力机制，学习与方面项相关的局部语义特征，再结合自注意力机制学习句子的全局语义特征。其次，为了更充分地利用依赖树中的句法信息，设计了利用依赖关系类型构建类型感知图模块，并采用注意力机制区分不同依赖类型的重要程度，重构带有权重的类型感知图。最后，通过图神经网络来挖掘更深层次的语义和句法信息。在Restaurant14、Laptop14和Twitter公开数据集上进行实验，实验结果表明，与基准模型相比，本文提出的模型具有更好的分类效果。 |
| 英文摘要: |
| In order to address the problems of underutilization of syntactic information in dependency trees and the inadequate extraction of semantic information in aspect-based sentiment analysis(ABSA) tasks, a dual graph convolutional network model based on hybrid attention and type-aware is proposed. First, a hybrid attention module is designed for more comprehensive extraction of semantic information of sentences, which employs an aspect-aware attention mechanism to learn local semantic features related to aspect terms, and then is combined with a self-attention mechanism to learn global semantic features of sentences. Secondly, in order to better utilize the syntactic information in the dependency tree, the module for constructing a type-aware graph based on dependency types is designed, and the attention mechanism is used to distinguish the importance of different dependency types and reconstruct the type-aware graph by assigning weights to the edges. Finally, a graph neural network is used to extract deeper semantic and syntactic information. Experiments are conducted on Restaurant14, Laptop14 and Twitter public datasets, and the experimental results show that the proposed model in this paper has better classification results compared with the baseline models. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
