Aspect Based Sentiment Analysis Model Based on Multi-Channel Feature Fusion

EasyChair Preprint 13044

Abstract

Aspect Based Sentiment Analysis (ABSA) is a natural language processing (NLP) task used for recognizing sentiment polarity about given aspects in the sentence. The current methods mostly combine the semantics and syntax of text. Which disregards the assistance of integrating external knowledge to enhance feature representation. In this context, the paper introduces a Multi-Channel Feature Fusion (MCFF) model, which captures sentiment feature from three diverse perspectives: context-based, syntax-based and aspect-term-concept-enhanced. Firstly, the model proposed in this paper learns context and syntax representations in parallel to adequately extract semantic features. Secondly, after enhancing aspect terms with external knowledge, the model merges them with specific aspect features to obtain knowledge-enhanced aspect representations. We extract sentiment features from multiple perspectives and fuse these feature representations to acquire the final text representation. The model in this paper has been verified to have excellent performance on standard datasets.

Keyphrases: Concept Enhancement, Graph Convolutional Network, aspect-based sentiment analysis, feature fusion

@booklet{EasyChair:13044,
  author    = {Hao Wang and Zhonglin Zhang},
  title     = {Aspect Based Sentiment Analysis Model Based on Multi-Channel Feature Fusion},
  howpublished = {EasyChair Preprint 13044},
  year      = {EasyChair, 2024}}