The authors propose CAVE-ABSA, a Constraint-Aware Validated Editing framework designed to generate and validate aspect-level counterfactuals for Aspect-Based Sentiment Analysis (ABSA). This approach addresses the challenge of flipping sentiment for a specific target aspect while preserving the meaning and sentiment of non-target aspects.
CAVE-ABSA localizes opinion spans, performs controlled rewriting, and refines candidates through a repair module. It filters outputs using aspect-level verification, semantic similarity, AMR-guided structural preservation, edit minimality, fluency, and contradiction detection.
The framework aims to construct validated counterfactual datasets for robustness evaluation and data augmentation, testing whether ABSA models rely on aspect-grounded sentiment reasoning.