UKPPsycontrol at SemEval-2026 Task 2: Modeling Valence and Arousal Dynamics from Text

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

Summary

The UKPPsycontrol system, presented at SemEval-2026 Task 2, addresses the challenge of modeling both current affect and short-term affective changes within chronologically ordered user-generated texts. Researchers Darya Hryhoryeva, Amaia Zurinaga, Hamidreza Jamalabadi, and Iryna Gurevych explored three distinct approaches: LLM prompting in both user-aware and user-agnostic configurations, a pairwise Maximum Entropy (MaxEnt) model incorporating Ising-style interactions for structured transition modeling, and a lightweight neural regression model that integrates recent affective trajectories and trainable user embeddings. Key findings from this work, published in the Proceedings of the 20th International Workshop on Semantic Evaluation (2026) on pages 528–539, indicate that while Large Language Models effectively capture static affective signals from text, short-term affective variation in the dataset is more accurately explained by recent numeric state trajectories rather than textual semantics alone.

Key takeaway

For NLP Engineers developing systems to model affect dynamics from user-generated text, recognize that while Large Language Models effectively capture static emotional states, you should prioritize incorporating recent numeric state trajectories over purely textual semantics for accurately predicting short-term affective changes. Your models will benefit from integrating user embeddings to capture individual affective patterns.

Key insights

Short-term affective variation in text is better explained by numeric state trajectories than textual semantics.

Principles

• Numeric state trajectories explain affective variation.
• User embeddings can enhance affect modeling.
• LLMs excel at static affective signal capture.

Method

The system combines LLM prompting (user-aware/agnostic), a pairwise Maximum Entropy model with Ising-style interactions, and a lightweight neural regression model using affective trajectories and user embeddings.

In practice

• Prioritize numeric state for dynamic affect modeling.
• Employ LLMs for static affect detection.
• Integrate trainable user embeddings.

Topics

• Affective Computing
• Valence-Arousal Modeling
• LLM Prompting
• Maximum Entropy Models
• Neural Regression
• User Embeddings

Best for: Research Scientist, AI Scientist, 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.
• Multimodal Hidden Markov Models for Persistent Emotional State Tracking — Sticky factorial HDP-HMMs effectively track persistent emotional regimes in multimodal conversations, outperforming standard HMMs and enhancing LLM responses.
• Emotion-Attended Stateful Memory (EASM):The Architecture for Hyper-Personalization at Scale — Emotion-attended stateful memory significantly enhances language model personalization by integrating user history and emotional signals.
• 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.
• Do We Still Need Humans in the Loop? Comparing Human and LLM Annotation in Active Learning for Hostility Detection — LLM annotation at scale matches human aggregate performance at lower cost, but with distinct error profiles.
• Emotional Modulation in Swarm Decision Dynamics — Emotional valence and arousal significantly modulate collective decision dynamics by altering recruitment and inhibition rates.

Open in AIssential →

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