YNU-HPCC at SemEval-2026 Task 2: Contrastive Calibration and Temporal Modeling for Continuous Valence-Arousal Prediction

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

Summary

The YNU-HPCC system achieved 2nd place in both subtasks of SemEval-2026 Task 2, focusing on continuous valence-arousal prediction. For Subtask 1, which addresses static affect state estimation and aims to mitigate semantic ambiguity, the team developed a hard-prompt-based regression model. This model was enhanced with unsupervised contrastive learning (SimCSE) and supervised contrastive calibration (SCL), leveraging an external affect lexicon to improve textual representation consistency and scale stability in the Valence–Arousal (V/A) space. For Subtask 2a, involving dynamic change prediction with irregular time intervals and historical dependencies, a Time-Aware LSTM architecture was introduced. This architecture integrates current affective states with temporally enriched historical trajectories. The system scored 0.677 for Valence and 0.528 for Arousal in Subtask 1, and 0.692 for Valence and 0.647 for Arousal in Subtask 2a.

Key takeaway

For NLP engineers developing continuous affect prediction systems, consider integrating contrastive learning and temporal modeling. Your static state models can benefit from hard-prompt regression enhanced with unsupervised SimCSE and supervised contrastive calibration, especially when grounded in external affect lexicons. For dynamic predictions, a Time-Aware LSTM architecture effectively incorporates historical dependencies, improving accuracy in irregular time series. This approach can significantly boost your system's performance in complex affective computing tasks.

Key insights

Contrastive calibration and temporal modeling significantly improve continuous valence-arousal prediction performance.

Principles

• Mitigate semantic ambiguity with contrastive learning.
• Integrate historical data for dynamic affect prediction.
• Lexicon-grounded calibration enhances V/A consistency.

Method

A hard-prompt-based regression model with SimCSE and SCL for static states, and a Time-Aware LSTM for dynamic, temporally enriched historical trajectories.

In practice

• Apply SimCSE/SCL for V/A representation.
• Use Time-Aware LSTMs for time-series affect.
• Incorporate external affect lexicons.

Topics

• Affective Computing
• Valence-Arousal Prediction
• SemEval-2026 Task 2
• Contrastive Learning
• Time-Aware LSTM
• Natural Language Processing

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

Related on AIssential

• 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.
• 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.
• The Circumplex Degeneracy Behind the Rare-Class Limit in Affect Recognition — Rare emotion recognition failures stem from affect geometry's circumplex degeneracy, not just class imbalance, demanding new distinguishing representations.
• Training-Free Test-Time Contrastive Learning for Large Language Models — Frozen LLMs can self-improve online by distilling contrastive supervision from their own inference experiences without gradient updates or external knowledge.
• Observable Neural ODEs for Identifiable Causal Forecasting in Continuous Time — Observability of latent states is crucial for identifying dynamic treatment effects in continuous-time causal forecasting with hidden confounders.
• Deep Temporal Modeling and Ensemble Fusion for Multimodal Emotion Recognition from Physiological Signals — Combining deep temporal models with sensor and ensemble fusion significantly boosts physiological emotion recognition accuracy.

Open in AIssential →

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