Aspect Sentiment Classification via Local Context-Focused Syntax Based on DeBERTa
Jie Liu, Zhenguo Zhang, Xinghua Lu · 2024
Aspect-Based Sentiment Analysis (ABSA) aims to predict the sentiment polarity of different aspects within sentences or documents. ABSA comprises four main tasks: Aspect Term Extraction (ATE), Aspect Category Detection (ACD), Opinion Term Extraction (OTE), and Aspect Sentiment Classification (ASC), with ASC receiving significant attention and achieving commendable results. In recent years, pre-trained models (PTMs) have been widely employed to address ASC challenges, yet whether PTMs encapsulate sufficient syntactic information for ASC necessitates further research. In this paper, we first explore the DeBERTa model (BERT with disentangled attention mechanism and enhanced mask decoder) to tackle ASC problems. Then, to further focus on the syntactic relationships between aspects, we combine the Multi-Head Self-Attention (MHSA) based Local Context Focus (LCF) with syntax, forming the Local Context Focus Syntax (LCFS) analysis for sentiment classification. By integrating the DeBERTa model, we propose a classification model based on Local Context Focus Syntax (LCFS) analysis. Experimental results demonstrate promising performance using DeBERTa with LCFS analysis on ASC tasks in the automotive domain.