基于图卷积网络的方面水平情感分类
Syntactic Tree-Enhanced GCN for Aspect-Level Sentiment Classification
摘要
针对不同方面的情感分类，其目的是判别评论语句对不同意见目标的情感极性。当一个句子包含多个[意见目标-情感]关联对时，该任务的主要挑战是为不同的意见目标匹配对应的意见文本以及情感极性。同时方面水平情感分类的数据标注工作较为繁杂，这导致现有数据集规模非常小，这导致模型训练的过程中会产生过拟合问题，严重影响了实验结果。本文建立了一个新的基于句法树的图卷积网络（GCN）来建立不同方面词与对应上下文的语义信息关联，GCN与句法树的组合也被证明能够较好地关联长距离非近邻词。基于SemEval 2014数据集的实验结果表明，与现有方法相比，本文提出的模型更具竞争力。
Abstract
Aspect-level sentiment classification aims to identify the sentiment polarity of review sentences toward different opinion targets. When a single sentence contains multiple pairs of opinion targets and corresponding sentiments, the core challenge of this task lies in matching each opinion target with its corresponding opinion context and correct sentiment polarity. In addition, the elaborate labeling required for aspect-level sentiment analysis results in small-scale public datasets, which easily triggers severe overfitting during model training and degrades experimental performance. To address these issues, this paper proposes a novel syntactic tree-based Graph Convolutional Network (GCN) to explicitly model semantic connections between aspect terms and their corresponding contexts. The combination of GCN and syntactic trees is verified to effectively model long-distance non-adjacent word dependencies. Experiments conducted on the SemEval 2014 dataset demonstrate that the proposed model achieves competitive performance compared with state-of-the-art baseline methods.关键词
人工智能/图卷积网络/情感分类/句法路径/自注意机制Key words
Artificial Intelligence/Graph Convolutional Network/sentiment classification/syntactic path/self-attention mechanism引用本文复制引用
王全举.基于图卷积网络的方面水平情感分类[EB/OL].(2026-10-08)[2026-10-09].http://www.paper.edu.cn/releasepaper/content/202610-1.学科分类
计算技术、计算机技术
