Aspect-Based Sentiment Analysis in Korean Based on Hierarchical Masking and Aspect-aware Attention

Aspect-based Sentiment Analysis (ABSA) is considered a pivotal component of fine-grained sentiment analysis, focusing on predicting the sentiment polarity of a specific aspect within a given context. However, existing models for aspect-level sentiment analysis often fall short in seamlessly integrating the local semantic details of aspect words with the broader semantic context. In response, we introduce a novel model called Hierar-chical Masking and Aspect-aware Attention (HMAA). This model aims to depict nuanced relationships between opinion words and aspects, while also considering the overall sentiment tendencies of sentences.HMAA incorporates a hierarchical masking module that emphasizes diverse combinations of aspect words and their surrounding context. Additionally, an aspect-aware attention module is devised to comprehensively grasp the connections between aspect words and sentences on a global scale. The local context semantics collaboratively interact with the aspect-aware attention network for the final sentiment classification. Experimental validation on multiple Korean-language datasets demonstrates the efficacy of the proposed model.