EcoAffectTrack at SemEval-2026 Task 2: A Hierarchical DeBERTa-Transformer Framework with CCC Optimization for Longitudinal Affect Modeling

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

Summary

EcoAffectTrack, a hierarchical framework presented at SemEval-2026 Task 2, addresses longitudinal affect modeling by predicting emotional valence and arousal variations over time. The system integrates a DeBERTa-v3 encoder backbone, optimized with a differentiable Concordance Correlation Coefficient (CCC) Loss for affect assessment (Subtask 1). This optimization prioritizes capturing the "shape" and trend of emotional trajectories, demonstrating a significant performance gain compared to standard Mean Squared Error. For state change forecasting (Subtask 2A), the framework incorporates a Transformer-based temporal forecaster, utilizing positional encoding to manage inter-subject emotional baseline variability. Additionally, disposition profiling (Subtask 2B) is handled by a deep attention network that aggregates historical embeddings to identify emotionally informative essays. Experimental results emphasize the importance of aligning loss functions with evaluation metrics and employing task-specific temporal modeling for robust longitudinal emotion recognition.

Key takeaway

For Machine Learning Engineers developing longitudinal affect models, prioritizing the "shape" of emotional trajectories over absolute point-wise accuracy is crucial. You should align your loss functions, such as the Concordance Correlation Coefficient (CCC) Loss, directly with evaluation metrics to achieve significant performance gains. Additionally, integrate task-specific temporal modeling, like Transformer-based forecasters with positional encoding, to robustly account for inter-subject variability in emotional baselines.

Key insights

Effective longitudinal affect modeling requires aligning loss functions with evaluation metrics and utilizing task-specific temporal approaches.

Principles

• Prioritize emotional trajectory "shape" over point-wise accuracy.
• Align loss functions directly with evaluation metrics.
• Employ task-specific temporal modeling for robustness.

Method

The framework uses a DeBERTa-v3 encoder with CCC Loss for affect assessment, a Transformer-based temporal forecaster for state change, and a deep attention network for disposition profiling, all within a hierarchical structure.

In practice

• Apply CCC Loss for affect assessment to capture trends.
• Integrate positional encoding in temporal forecasters.
• Aggregate historical embeddings for disposition profiling.

Topics

• Longitudinal Affect Modeling
• DeBERTa-v3
• Transformer Networks
• CCC Loss
• Emotion Recognition
• SemEval-2026

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

Related on AIssential

• 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.
• 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.
• EEGDancer: Dynamic Emotion Latent Space Masked Modeling with Reinforcement Learning for EEG Continuous Emotion Prediction — Continuous EEG emotion prediction benefits from a unified framework combining discrete latent space learning, masked temporal modeling, and reinforcement learning.
• Two-Stage Multimodal Framework for Emotion Mimicry Intensity Prediction — Staged multimodal fusion with independent encoder pre-training improves robustness for continuous emotion intensity prediction.
• Team MKC at CLPsych 2026: Capturing and Characterizing Mental Health Changes through Social Media Timeline Dynamics — An LLM-based pipeline analyzes social media timelines for early mental health change detection and continuous monitoring, integrating post and user-level assessments.
• From Correlation to Cause: A Five-Stage Methodology for Feature Analysis in Transformer Language Models — The five-stage methodology links correlation to causation in Transformer feature analysis, revealing partial causality and robustness gaps.

Open in AIssential →

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