looploop at SemEval-2026 Task 3: A Dimensional Aspect-Based Sentiment System with DeBERTa Regression and Qwen3 Instruction Fine-Tuning

AbstractAspect-Based Sentiment Analysis (ABSA) hasevolved to capture continuous affective states,posing challenges for traditional classificationmodels. We adopt a hybrid approach tailoredto the varying complexities of the subtasks. ForTask 1 (Valence-Arousal Regression), we em-ploy a discriminative architecture using pre-trained DeBERTa encoder with a MeanPool-ing mechanism to directly regress continuoussentiment scores. For Tasks 2 and 3, which re-quire complex structural extraction of opiniontriplets and quadruplets, we utilize a generativeapproach by fine-tuning the Qwen3-4B-Instructlarge language model via 4-bit QLoRA. Oursystem effectively handles both precise numer-ical regression and complex structural text gen-eration, achieving competitive results acrossthe English laptop and restaurant domains.