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

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

The looploop system, developed for SemEval-2026 Task 3, presents a hybrid approach to Aspect-Based Sentiment Analysis (ABSA) that addresses the challenges of capturing continuous affective states. For Task 1, which involves Valence-Arousal Regression, the system employs a discriminative architecture utilizing a pre-trained DeBERTa encoder with a MeanPooling mechanism to directly regress continuous sentiment scores. For the more complex Tasks 2 and 3, requiring structural extraction of opinion triplets and quadruplets, a generative approach is adopted. This involves fine-tuning the Qwen3-4B-Instruct large language model using 4-bit QLoRA. This dual strategy effectively manages both precise numerical regression and intricate structural text generation, yielding competitive results across English laptop and restaurant domains.

Key takeaway

For NLP engineers developing advanced sentiment analysis systems, consider adopting a hybrid modeling approach. Your strategy should involve discriminative models like DeBERTa for precise continuous sentiment regression and instruction-tuned LLMs such as Qwen3-4B-Instruct for complex structural information extraction. This allows you to optimize performance across diverse ABSA subtasks, leveraging 4-bit QLoRA for efficient LLM deployment.

Key insights

A hybrid ABSA system combines discriminative regression with generative instruction fine-tuning for varying task complexities.

Principles

• Match model architecture to task complexity.
• Pre-trained encoders excel at continuous regression.
• LLMs are effective for complex structural generation.

Method

For Task 1 (Valence-Arousal Regression), use DeBERTa with MeanPooling. For Tasks 2 and 3 (structural extraction), fine-tune Qwen3-4B-Instruct via 4-bit QLoRA.

In practice

• Apply DeBERTa for continuous sentiment scoring.
• Utilize Qwen3-4B-Instruct for opinion triplet extraction.
• Employ 4-bit QLoRA for efficient LLM fine-tuning.

Topics

• Aspect-Based Sentiment Analysis
• DeBERTa
• Qwen3-4B-Instruct
• QLoRA
• Sentiment Regression
• Instruction Fine-Tuning
• SemEval-2026

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Can Machines Tell Whether Gamers Love or Hate a Game? — Fine-tuning Transformer models like RoBERTa significantly improves sentiment analysis accuracy on imbalanced, complex human language datasets.
• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific LoRA fine-tuning on Qwen2.5-14B with CCPoetry-49K significantly enhances classical Chinese poetry understanding and translation.
• Portuguese Sentiment Analysis with Open-Source LLMs: Models, Prompts, and Efficient Deployment — Open-source LLMs offer competitive Portuguese sentiment analysis with efficient deployment, especially using CoT+FS prompting.
• DiffuSent: Towards a Unified Diffusion Framework for Aspect-Based Sentiment Analysis — DiffuSent unifies ABSA subtasks via non-auto-regressive boundary denoising, improving accuracy and speed for multi-word sentiment extraction.
• Aspect-Based Sentiment Analysis in the Accommodation Domain Using the BERTimbau Model — BERTimbau model fine-tuning strategies for Portuguese ABSA show trade-offs between recall and precision.
• Sentiment Analysis with BERT: The Heart of the Transformer, The Soul of the Text — BERT's bidirectional context understanding via Transformers significantly advanced sentiment analysis and NLP tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.