The paper introduces a novel data augmentation framework, LASS, combining large and small language models to alleviate emotional bias in opinion summarization. By rewriting positive text into negative reviews and training a disentangle reconstruction model, the framework generates synthetic data effectively. Experimental results show improved sentiment accuracy without compromising summary quality.
לשפה אחרת
מתוכן המקור
arxiv.org
תובנות מפתח מזוקקות מ:
by Yanyue Zhang... ב- arxiv.org 03-13-2024
https://arxiv.org/pdf/2403.07693.pdfשאלות מעמיקות