Khaleesiyali at SemEval-2026 Task 2: Lexicon-Augmented RoBERTa for Valence–Arousal Regression on Ecological Essays
Year: 2026
Type: conference-paper
Abstract: This paper presents a lexicon-augmented RoBERTa system for the SemEval-2026 Task 2 valence-arousal regression challenge.The model integrates deep contextual embeddings with a 6-dimensional feature vector derived from the NRC VAD lexicon, achieving a high token coverage rate of 72.05%.Under official user-aware evaluation, the system reached a competitive average composite correlation of 0.547, significantly outperforming the ridgeregression baseline.The system demonstrated particular robustness in valence (r = 0.656) and achieved strong generalization to unseen users (r arousal = 0.519).These findings indicate that lightweight lexicon-based statistics provide valuable complementary cues for longitudinal emotion modeling in modern transformer architectures.
Author: Eleale Tee
Institution: Waseda University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166819832