首页|基于MHSA和GCN的方面级情感分析模型

基于MHSA和GCN的方面级情感分析模型

扫码查看
针对目前大多数现有的基于图卷积网络的模型只考虑了特定方面和上下文之间的交互关系,忽略了方面之间的交互情感特征的问题,本文提出了一种利用预训练BERT和多头自注意力机制(MHSA)结合图卷积网络的模型(MHSAGCN-BERT).用方面词与上下文的句法依赖和方面之间的相互情感关系来推导出特定方面的情感极性,以此增强模型学习特征能力.在Restaurant14、Restaurant15、Restaurant16公开数据集上进行了实验,结果表明,本文模型与其他方面级情感分析模型相比有较明显的提升.
Aspect level sentiment analysis model based on MHSA and GCN
Aspect-level sentiment analysis is a fine-grained sentiment analysis task.Most of the existing graph convolutional network-based models only consider the interaction between specific aspects and contexts,largely ignoring the interactive emotional features between aspects.Aiming at this deficiency,a model (MHSAGCN-BERT) that utilizes pre-trained BERT and a Multi-head Self-attention mechanism (MHSA) combined with a graph convolutional network is proposed.The syntactic dependencies of aspect words and context and the mutual sentiment relationship between aspects are used to derive aspect-specific sentiment polarity,thereby enhancing the model's ability to learn features.The experimental results on three public datasets of international semantic evaluation show that the model is significantly improved compared with other aspect-level sentiment analysis models.

aspect-level sentiment analysismulti-head self-attention mechanismgraph convolutional networksaspect interactionsyntactic dependency tree

杨乾、艾山·吾买尔、孙伟伟、古文霞

展开 >

新疆大学信息科学与工程学院,新疆 乌鲁木齐 830046

新疆大学新疆多语种信息技术实验室,新疆 乌鲁木齐 830046

方面级情感分析 多头自注意力机制 图卷积网络 方面交互 句法依赖树

新疆乌鲁木齐自治区自然科学基金联合基金面上项目国家重点研发项目子课题项目天山创新团队计划项目

2021D01C0812018YFB14032022020D14044

2024

东北师大学报(自然科学版)
东北师范大学

东北师大学报(自然科学版)

CSTPCD北大核心
影响因子:0.612
ISSN:1000-1832
年,卷(期):2024.56(2)
  • 16