Stochastic Gradient Descenders at SemEval-2026 Task 9: Few-Shot LLM Prompting for Polarization Type Classification

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

Summary

A system developed by Huynh Phu and Dang Thin for SemEval-2026 Task 9 (POLAR), Subtask 2, addresses the classification of polarization types in social media text. The researchers investigated three distinct paradigms: fine-tuning mDeBERTa-v3 with domain-adaptive pre-training, parameter-efficient adaptation of Qwen2.5-32B using LoRA, and few-shot prompting with Llama-3.3-70B-Instruct. Experimental results demonstrated that few-shot prompting, utilizing Llama-3.3-70B-Instruct, significantly outperformed both fine-tuning and parameter-efficient approaches, despite requiring no task-specific training. This method achieved non-zero F1 scores across all polarization categories, which is crucial for macro-averaged evaluation. The system secured 2nd place among 29 English submissions on the official leaderboard, with an F1 Macro of 0.5157, underscoring the efficacy of large instruction-tuned models in low-resource, label-imbalanced classification scenarios.

Key takeaway

For NLP Engineers classifying social media text in low-resource or label-imbalanced scenarios, you should prioritize few-shot prompting with large instruction-tuned models like Llama-3.3-70B-Instruct. This approach demonstrated superior performance over fine-tuning mDeBERTa-v3 and LoRA adaptation of Qwen2.5-32B, securing a 2nd place ranking at SemEval-2026 Task 9. Evaluate few-shot prompting early in your development cycle to achieve competitive results with minimal task-specific training.

Key insights

Few-shot prompting with Llama-3.3-70B-Instruct outperformed fine-tuning and LoRA for polarization classification, ranking 2nd in SemEval-2026 Task 9.

Principles

• Few-shot prompting can outperform fine-tuning.
• Instruction-tuned LLMs excel in low-resource settings.
• Macro-averaged F1 requires non-zero category scores.

Method

The system employed few-shot prompting with Llama-3.3-70B-Instruct, comparing it against mDeBERTa-v3 fine-tuning and Qwen2.5-32B LoRA adaptation for social media polarization classification.

In practice

• Use Llama-3.3-70B-Instruct for few-shot tasks.
• Prioritize instruction-tuned LLMs for imbalanced data.
• Evaluate few-shot prompting before fine-tuning.

Topics

• Few-Shot Prompting
• Large Language Models
• Polarization Classification
• SemEval-2026
• Instruction Tuning
• Low-Resource NLP

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

Related on AIssential

• Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification — In-context learning for Turkish LVCs is highly prompt-sensitive, with few-shot demonstrations improving LLM calibration to match supervised baselines.
• AI 101: Beyond RL: The New Fine-Tuning Stack for LLMs — Modern LLM fine-tuning is evolving into a modular stack, moving beyond monolithic RL to integrate diverse, efficient methods.
• Scikit-LLM vs. Traditional Text Classifiers: When Should You Use an LLM? — LLMs offer superior zero-shot text classification accuracy and competitive latency, especially with limited data.
• MAF: Multimodal Adaptive Few-shot Prompting for Sentiment Analysis with MLLMs — MAF dynamically adapts few-shot prompts for MLLMs in sentiment analysis by integrating multimodal cues and adaptive demonstration retrieval.
• How Small Can You Go? LoRA Fine-Tuning 270M-8B Models for Merchant Information Extraction in Financial Transactions — Smaller LoRA-fine-tuned models can achieve near 8B performance for structured extraction, significantly reducing deployment costs.
• Soft-Prompt Tuning for Fair and Efficient LLM Benchmark Evaluation — Soft-prompt tuning efficiently reveals true LLM knowledge by overcoming format-following limitations in benchmarks.

Open in AIssential →

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