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

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

Summary

SRCB's system for the SemEval-2026 Task 3 DimASR subtask addresses dimensional sentiment regression, specifically predicting Valence-Arousal scores from English restaurant reviews. The core of their approach combines Qwen3 large language models with a novel contrastive LLM-based data augmentation framework. This augmentation strategy is designed to enrich the training data, allowing the system to capture subtle affective variations and nuanced shifts at the aspect level more effectively. Experimental results confirm that this data augmentation significantly improves overall performance on the DimASR task. The system achieved a competitive score of 1.227 RMSE on the test set, demonstrating the framework's efficacy in complex sentiment analysis challenges.

Key takeaway

For Machine Learning Engineers developing sentiment analysis models, especially those struggling with subtle affective shifts, consider integrating contrastive LLM-based data augmentation. This approach, demonstrated with Qwen3 LLMs for Valence-Arousal regression, significantly enhances a model's ability to capture fine-grained sentiment at the aspect level. You should explore similar augmentation strategies to improve performance on complex dimensional sentiment tasks.

Key insights

Contrastive LLM-based data augmentation significantly boosts dimensional sentiment regression performance, especially for subtle aspect-level affective shifts.

Principles

• LLM-based augmentation can enrich training data.
• Contrastive learning enhances affective variation capture.
• Aspect-level shifts benefit from targeted data enrichment.

Method

The system uses Qwen3 LLMs with contrastive data augmentation to generate diverse training examples, improving Valence-Arousal score prediction in restaurant reviews.

In practice

• Apply contrastive LLM augmentation for sentiment tasks.
• Use Qwen3 LLMs for data generation.
• Target aspect-level sentiment with augmented data.

Topics

• SemEval-2026 Task 3
• Dimensional Sentiment Regression
• Valence-Arousal Scores
• Qwen3 LLMs
• Contrastive Data Augmentation
• Aspect-Level Sentiment

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

Related on AIssential

• Beyond Sentiment Classification: A Generative Framework for Emotion Intensity Evaluation in Text — Generative LLMs fine-tuned on continuous emotion intensity scores outperform classification and generalize to unseen affective dimensions.
• mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection — Data augmentation and self-training can effectively finetune LLMs for specialized text classification tasks with limited data.
• LLM-Assisted Reranking to Operationalize Nuanced Objectives in Recommender Systems — LLM-assisted reranking can amplify extreme content, but prompt regularization can mitigate this while maintaining personalization.
• Multi-View Attention Multiple-Instance Learning Enhanced by LLM Reasoning for Cognitive Distortion Detection — Combining LLMs with MIL and ELB components improves cognitive distortion detection by enhancing interpretability and reasoning.
• Is Sentiment Analysis in Qualitative Data Analysis Software Accurate? — Sentiment analysis accuracy in QDA tools varies significantly, with LLMs offering superior performance at higher costs.
• Better Literary Translation: A Multi-Aspect Data Generation and LLM Training Approach — Multi-aspect iterative refinement generates superior literary translation data for LLM training.

Open in AIssential →

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