A Cross-Domain Generative Data Augmentation Framework for Aspect-Based Sentiment Analysis

Aspect-based sentiment analysis (ABSA) is a crucial fine-grained task that aims to determine polarity in specific aspect term. Recent research has advanced prediction accuracy by pre-training models on ABSA tasks. However, due the lack of data, those cannot be trained effectively. In this paper, we propose cross-domain generative data augmentation framework (CDGDA) utilizes generation model produce in-domain, sentences learning from similar, coarse-grained datasets out-of-domain. To generate sentences, guide using two prompt methods: replacement and aspect–sentiment pair replacement. We also refine quality generated an entropy minimization filter. Experimental results three public show our outperforms most baseline methods other methods, thereby demonstrating its efficacy.