ES4MLL at SemEval-2026 Task 2: Set Attention Aggregation and Recurrent Temporal Modeling for Longitudinal Affect Prediction

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

Summary

ES4MLL presents a neural architecture for longitudinal affect prediction, developed for SemEval-2026 Task 2. This task utilizes a dataset of essays and feeling words annotated with self-reported valence and arousal scores. The proposed system combines pretrained Transformer encoders with temporal sequence modeling to predict continuous valence and arousal over user-specific timelines. It encodes individual texts using a Transformer-based language model, aggregates them via attention-based pooling, and then processes them with recurrent layers to capture longitudinal dependencies. The work also explores parameter-efficient fine-tuning strategies to adapt pretrained representations effectively under limited data conditions.

Key takeaway

For NLP engineers developing longitudinal affect prediction systems, you should consider integrating Transformer encoders with recurrent temporal modeling. This approach, especially when combined with parameter-efficient fine-tuning, can effectively capture both linguistic content and temporal emotional dynamics. It allows for accurate prediction of continuous valence and arousal over user-specific timelines, even when working with limited annotated data, enhancing the robustness of your models.

Key insights

Longitudinal affect prediction effectively combines Transformer encoders with recurrent temporal modeling for user-specific timelines.

Principles

• Longitudinal affect prediction requires capturing both linguistic content and temporal emotional dynamics.
• Attention-based pooling can aggregate individual text encodings effectively.

Method

Encode individual texts using a Transformer-based language model, aggregate through attention-based pooling, then process with recurrent layers to capture longitudinal dependencies.

In practice

• Employ parameter-efficient fine-tuning for adapting pretrained models with limited data.
• Utilize self-reported valence and arousal scores for affect annotation.

Topics

• Longitudinal Affect Prediction
• SemEval-2026 Task 2
• Transformer Encoders
• Temporal Sequence Modeling
• Attention Aggregation
• Parameter-Efficient Fine-Tuning

Code references

• AndreaLolli2912/SemEval2026-EmoVA

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.
• Understanding Transformers, the MLE Way — Transformers use attention to process sequences non-sequentially, enabling faster and more efficient deep learning for various tasks.
• 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.
• EmoNet: Speaker-Aware Transformers for Emotion Recognition — and What I’d Build Differently in 2026 — Speaker-specific patterns and historical context are crucial for accurate emotion recognition in conversations.
• Emotion-Aware Image Generation from Korean Diary Text via LLM-based Prompt Translation and LoRA Fine-Tuning — A pipeline generates emotion-aware, children's drawing style images from Korean diary text using LLM sentiment analysis and LoRA fine-tuned Stable Diffusion.
• 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.

Open in AIssential →

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