计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (5): 148-155.DOI: 10.3778/j.issn.1002-8331.2009-0174

• 模式识别与人工智能 • 上一篇    下一篇

结合语法信息的BG-CNN用于方面级情感分类

郑诚,魏素华,曹源   

  1. 安徽大学 计算机科学与技术学院,合肥 230601
  • 出版日期:2022-03-01 发布日期:2022-03-01

BG-CNN Combined with Grammatical Information for Aspect Level Sentiment Classification

ZHENG Cheng, WEI Suhua, CAO Yuan   

  1. School of Computer Science and Technology, Anhui University, Hefei 230601, China
  • Online:2022-03-01 Published:2022-03-01

摘要: 方面级别的情感分析(ABSA)旨在确定句子中特定目标的情感倾向。大部分现有方法仅使用语义层面信息,不能很好地利用不同方面词的意见术语来达到精确的情感分类,且模型不具有可解释性。语法层面信息中词性信息和以特定方面术语为根节点的句法结构依存树可以用于捕获句子中特定方面的意见术语。提出了结合词性信息且具有模型可解释性的BG-CNN,并引入依存树作为辅助信息用于细粒度文本情感分析。提出了增强损失函数用于模型的训练。在三个经典数据集上进行验证,实验结果表明了该模型和增强损失函数的有效性。

关键词: 方面级情感分析, 依存树, 语法层面, 模型可解释性

Abstract: Aspect level sentiment analysis(ABSA) aims to determine the emotional orientation of a particular goal in a sentence.Most of the existing methods only use semantic information, and can’t make good use of the opinion terms of different aspects to achieve accurate emotion classification, and the model is not interpretable. Part of speech information in grammar level information and syntactic structure dependency tree with specific aspect words as root nodes can be used to capture opinion terms of specific aspects in sentences. Therefore, this paper proposes a BG-CNN with part of speech information and model interpretability, and introduces dependency tree as auxiliary information for fine-grained text sentiment analysis. In addition, an enhanced loss function is proposed to train the model. Experimental results on three classical datasets show the effectiveness of the model and the enhanced loss function.

Key words: aspect level sentiment analysis, dependency tree, grammatical level, model explicable