SRCB at SemEval-2026 Task 3: Boosting DimASR via Contrastive LLM-Based Data Augmentation

AbstractWe present our system for the DimASR subtask of SemEval-2026 Task 3: DimABSA, targeting dimensional sentiment regression of Valence-Arousal scores in English restaurant reviews. Our approach leverages Qwen3 large language models combined with contrastive LLM-based data augmentation to enrich training data and capture subtle affective variations. Experiments show that this data augmentation framework significantly improves performance on the DimASR task, particularly in capturing subtle affective shifts at the aspect level. Finally, our system achieves a score of 1.227 RMSE on the test set.