PAI at SemEval-2026 Task 3: An LLM and Data Redistribution Adaptation-Based Predictive Strategy for Valence-Arousal Scores

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

Summary

PAI's strategy for SemEval-2026 Task 3 addresses the valence and arousal score prediction task within Dimensional Aspect-Based Sentiment Analysis (DimABSA). The proposed method employs a two-stage approach. Initially, a Large Language Model (LLM) undergoes post-training through a Supervised Fine-Tuning (SFT) scheme to produce preliminary valence and arousal score predictions. Subsequently, the second stage focuses on adapting the distribution of these initial results. This adaptation process leverages the training set's distribution, incorporating Gaussian distribution modeling, quantile mapping, and the Sinkhorn algorithm to refine final predictions. This comprehensive strategy aims to enhance the accuracy of sentiment dimension predictions.

Key takeaway

For NLP Engineers developing advanced sentiment analysis systems, particularly for Dimensional Aspect-Based Sentiment Analysis (DimABSA), you should consider a two-stage predictive strategy. This involves initially fine-tuning a Large Language Model (LLM) for core predictions, then applying distribution adaptation techniques like Gaussian modeling or quantile mapping to refine results based on training data characteristics. This approach can significantly improve the accuracy of valence and arousal score predictions.

Key insights

A two-stage strategy combines LLM fine-tuning with distribution adaptation for valence-arousal prediction in DimABSA.

Method

The method involves two stages: first, post-training an LLM via Supervised Fine-Tuning (SFT) for initial valence/arousal scores; second, adapting these scores' distribution using Gaussian modeling, quantile mapping, and the Sinkhorn algorithm.

Topics

• Large Language Models
• Supervised Fine-Tuning
• Dimensional Sentiment Analysis
• Valence-Arousal Prediction
• Distribution Adaptation
• SemEval

Best for: Research Scientist, AI Scientist, 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.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• Consistency Analysis of Sentiment Predictions using Syntactic & Semantic Context Assessment Summarization (SSAS) — SSAS improves LLM sentiment prediction consistency by pre-processing noisy data into structured, high-signal prompts.
• Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity — LLMs, guided by a context-aware "Complexity Score" for prompt selection, significantly outperform fine-tuned models on inferentially complex, low-prevalence data extraction.
• 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.
• From Affect Prediction to Affect Forecasting: Evidence for Distinct Information Sources in Longitudinal Text — Current affect prediction relies on text semantics, while future affective change forecasting benefits from numeric trajectory dynamics.

Open in AIssential →

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