J. Wei

Label-guided data augmentation for cross-domain aspect-based sentiment analysis via enhanced affinity fusion

This work introduces an Enhanced Affinity Fusion module that explicitly strengthens aspect–opinion relational modeling by selectively integrating complementary attention mechanisms, and proposes Label-Guided Data Amplification (LGDA), which enhances supervision diversity and domain robustness through label-driven text expansion, hard sample mining, and domain-adaptive sampling.