计算机与数字工程2024，Vol.52Issue(4)：1068-1074,1124,8.DOI:10.3969/j.issn.1672-9722.2024.04.019
融合注意力机制的胶囊网络方面级情感分析
Joint Attention Mechanism and Capsule Network for Aspect-level Sentiment Analysis
李维乾 1李思雨1
作者信息
- 1. 西安工程大学计算机科学学院 西安 710600
- 折叠
摘要
Abstract
Aspect level sentiment analysis aims to clarify the emotional polarity of specific aspects in the text.Aiming at the problem that aspect words in sentences are composed of complex phrases,which leads to the wrong judgment of aspect emotion po-larity,this paper proposes a model based on attention mechanism and capsule network for aspect level sentient classification(AS-ATTCaps).The model first extracts the sequence semantic information through bi-directional long short term memory(BiL-STM),then encodes the target aspect in the sequence semantic information using N-gram model,and then uses the interactive at-tention mechanism to learn the attention between aspect words and context.The final generated text representation is connected to the capsule network of fusion aspect feature representation for classification,and the emotion classification results of text aspect lev-el are obtained.In this model,the capsule network is used to effectively extract the relationship between part and the whole,and the aspect feature transformation matrix extracted by N-gram model is integrated to improve the traditional dynamic routing method and enhance the judgment ability of the model on aspect emotional polarity.The experimental results show that the accuracy of the model on the two datasets reaches 78.4%and 72.4%,and the F1 scores are 0.687 and 0.668 respectively,which proves that the capsule network model with interactive attention mechanism has a strong classification effect in aspect level emotion analysis task.关键词
方面级情感分析/自然语言处理/胶囊网络/注意力机制Key words
aspect-based sentiment analysis/natural language processing/capsule network/attentional mechanism分类
信息技术与安全科学引用本文复制引用
李维乾,李思雨..融合注意力机制的胶囊网络方面级情感分析[J].计算机与数字工程,2024,52(4):1068-1074,1124,8.
