VerbaNex AI at SemEval-2026 Task 2: DeBERTa for Longitudinal Valence and Arousal Prediction

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

Summary

VerbaNex AI submitted two DeBERTa-based regression configurations to SemEval 2026 Subtask 1, "Longitudinal Affect Assessment," aiming to predict continuous valence and arousal from chronologically ordered texts. Their approach included a contextual model and a hybrid model integrating normalized lexical features from the NRC VAD lexicon. Both systems maintained temporal ordering and used user-level data splits to ensure generalization. The models achieved competitive performance, showing stronger results in valence prediction compared to arousal. Integrating lexical features did not consistently improve arousal scores, highlighting the difficulty in modeling emotional intensity dynamics. Error analysis revealed challenges with implicit emotions, pragmatic ambiguity, and subtle affective shifts over time.

Key takeaway

For NLP engineers developing longitudinal affect assessment systems, consider starting with DeBERTa-based models, prioritizing valence prediction where performance is stronger. You should implement user-level data splits to ensure model generalization to new individuals. Be aware that integrating lexical features might not consistently improve arousal prediction, indicating a need for more sophisticated approaches to capture emotional intensity dynamics over time.

Key insights

Combining contextual and lexical features improves longitudinal affect prediction, especially for valence.

Principles

• User-level data splits enhance model generalization.
• Temporal ordering is crucial for longitudinal assessment.
• Lexical features may not consistently aid arousal prediction.

Method

Fine-tune DeBERTa for regression, creating a contextual model and a hybrid model that incorporates normalized NRC VAD lexical features, preserving temporal order and using user-level data splits.

In practice

• Prioritize valence over arousal in initial models.
• Implement user-level data splits for robustness.
• Consider DeBERTa for longitudinal text analysis.

Topics

• Longitudinal Affect Assessment
• DeBERTa
• Valence Arousal Prediction
• SemEval 2026
• Lexical Features
• 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.
• JabuticaBERT: Modern Portuguese Encoders from Scratch with RTD and Long-Context Training — Monolingual encoders with careful architecture and tokenization can achieve competitive performance without massive scaling.
• 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.
• Machine learning and digital pragmatics: Which word category influences emoji use most? — MARBERT model shows promise in predicting Arabic emoji use, but needs further refinement for dialectal nuances.
• Aspect-Based Sentiment Analysis in the Accommodation Domain Using the BERTimbau Model — BERTimbau model fine-tuning strategies for Portuguese ABSA show trade-offs between recall and precision.
• 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.

Open in AIssential →

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