ASC-KEWSP: Aspect-level Sentiment Classification Model based on Knowledge-Enhanced and Word-Sentence-level Pre-training

Aspect-level Sentiment Classification (ASC) aims to determine the sentiment polarity of aspect words in a sentence, particularly in domains such as social media and e-commerce. Current methods mainly employ sequential representations or attention mechanisms to model the relationships between the context and target words, overlooking the connections at both the word and sentence levels, as well as the background knowledge of the text, resulting in insufficiently learned semantic relationships. In this paper, we propose an Aspect-level Sentiment Classification model named ASC-KEWSP, based on Knowledge-Enhanced, Word and Sentence-level Pre-training. This model combines external knowledge and word-sentencelevel pre-training tasks. External knowledge repositories introduce aspect words and knowledge vectors in different contexts to enhance the model’s understanding. Word-sentence-level pre-training enhances the model’s awareness of sentiment words through discriminator training. Experimental results demonstrate that, compared to existing models, ASC-KEWSP proposed in this paper is effective and reasonable for aspect-level sentiment classification tasks.