Joint Attention Mechanism and Capsule Network for Aspect-level Sentiment Analysis
LI Weiqian
LI Siyu
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.
Keywords：aspect-based sentiment analysisnatural language processingcapsule networkattentional mechanism
Publication Date：2024-04-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：8( 1068-1074,1124 )