节点文献
基于双通道胶囊网络特征融合的中文隐式情感分析
Dual-channel Capsule Network with Feature Fusion for Chinese Implicit Sentiment Analysis
【摘要】 文本情感分析作为自然语言处理的热门研究方向之一,在显式情感分析方面已取得了很多突破,而隐式情感方面的分析研究则相较缺乏。针对单一词向量输入无法充分表达文本语义的问题,该文采用CNN和BiLSTM混合神经网络提取文本的语义特征,同时将字、词、语义不同层级的特征通过双通道胶囊网络(Capsule Network)进行自主学习,随后输入交互注意力层进行融合。由实验结果可知,该文提出的模型在SMP2019_ECISA数据集上的准确率为84.83%,macro-F1值为82.76%,同时在对比实验中也取得了较好的效果,充分体现了该文模型的有效性。
【Abstract】 To address the Chinese implicit sentiment analysis,this paper uses CNN and BiLSTM hybrid neural network to extract the semantic features of the text instead of using single word vector input.At the same time,the features of characters,words and semantics are independently learned through the dual-channel Capsule Network,and then input to the interactive attention layer for fusion.The experimental results on SMP2019_ECISA data set show that the proposed model achieves 84.83% accuracy and 82.76% macro-F1value.
【Key words】 Chinese implicit sentiment analysis; dual-channel capsule network; multi-level feature fusion; RoBERTa;
- 【文献出处】 中文信息学报 ,Journal of Chinese Information Processing , 编辑部邮箱 ,2025年08期
- 【分类号】TP391.1
- 【下载频次】19