计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (18): 104-110.DOI: 10.3778/j.issn.1002-8331.1906-0418

• 模式识别与人工智能 • 上一篇    下一篇

用于特定目标情感分析的交互注意力网络模型

韩虎,刘国利   

  1. 1.兰州交通大学 电子与信息工程学院,兰州 730070
    2.甘肃省人工智能与图形图像工程研究中心,兰州 730070
  • 出版日期:2020-09-15 发布日期:2020-09-10

Interactive Attention Networks for Target-Based Sentiment Analysis

HAN Hu, LIU Guoli   

  1. 1.School of Electronic & Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
    2.Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphic & Image Processing, Lanzhou 730070, China
  • Online:2020-09-15 Published:2020-09-10

摘要:

特定目标情感分析旨在判别评论中不同目标所对应的情感极性。越来越多的研究人员采用基于神经网络的各种方法在特定目标情感分析任务中取得了较好的成绩。但大多数与目标相关的模型只关注目标对上下文建模的影响,而忽略了上下文在目标建模中的作用。为了解决上述问题,提出一种交互注意力网络模型(LT-T-TR),该模型将一条评论分为三个部分:包含目标的上文,目标,包含目标的下文。通过注意力机制进行目标与上下文的交互,学习各自的特征表示,从中捕获目标短语和上下文中最重要的情感特征信息。通过在两个标准数据集上的实验验证了模型的有效性。

关键词: 特定目标情感分析, 交互注意力网络, 注意力机制

Abstract:

Target-based sentiment analysis aims to identify the sentiment polarity of specific target in its context. In recent years, more and more researchers use various methods based on neural networks to solve target-based sentiment analysis, and achieve success. However, most target-related models only focus on the impact of target on context modeling, but ignore the role of context in target modeling. To address this problem, this paper proposes an interactive attention networks named LT-T-TR for target-based sentiment analysis, which divides a review into three parts:left context with target phrase, target phrase, and right context with target phrase. The interaction between target phrase and left/right context is utilized by attention mechanism to learn the representations of target phrase and left/right context separately. As a result, the most important words in target phrase or in left/right context are captured, and the results on laptop and restaurant datasets demonstrate the effectiveness of the model.

Key words: target-based sentiment analysis, interactive attention networks, attention mechanism