A Generative Model-Based Prediction of Aspect-Level Sentiment Quad with Stage Prompts and Semantic Enhancement

Increasing attention has been directed toward obtaining fine-grained sentiment analysis in natural language tasks, with quadruple extraction emerging as a key focal point. In the Aspect Sentiment Quad Prediction(ASQP) task, state-of-the-art generative approaches have achieved remarkable results. However, existing research has not yet sufficiently emphasized the importance of the relationships between these elements. We propose a novel generative model for tuples that focuses on recognizing all sentiment components in a sentence. The model is based on a pre-trained model and incorporates an AMR parser to parse sentences, prompt learning to generate tuples, and the use of cross-entropy to measure loss on the model. Comprehensive experimental results confirm the effectiveness of our approach and provide additional backing for the validity of our research motivation.