An External Knowledge-Guided Generative Model for Few-Shot Aspect Category Sentiment Analysis

Abstract

Few-shot Aspect Category Sentiment Analysis (ACSA) aims to predict the sentiment polarity of a given aspect category in scenarios with limited labeled data. A key challenge in ACSA is that aspect categories are often not explicitly mentioned in the text, requiring models to infer the relevant sentiment from context. Traditional classification-based approaches rely heavily on large labeled datasets and pre-trained knowledge, making them less effective in few-shot settings. To address these issues, we propose a generative ACSA model that incorporates external knowledge. We retrieve fine-grained aspect-related terms from external knowledge bases and use them to construct contrastive sentence pairs, generating enhanced aspect-related representations, thereby bridging the gap between the original text and predefined aspect categories. Additionally, we transform the classification task into a sequence generation process to predict sentiment polarity, aligning the pre-training task with the downstream objective and maximizing the use of pre-trained knowledge. Experimental results on three public datasets show that the proposed method significantly outperforms traditional classification models and existing generative models of few-shot ACSA, demonstrating its effectiveness.