@inproceedings{nguyen-2026-lamanhnguyen,
    title = "lamanhnguyen at {S}em{E}val-2026 Task 2: Uncovering Lexical Bias and Momentum Lag in Longitudinal Emotion Prediction using Multi-task {D}e{BERT}a",
    author = "Nguyen, Lam Anh",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.91/",
    pages = "630--634",
    ISBN = "979-8-89176-414-9",
    abstract = "This paper describes our system for SemEval-2026 Task 2: Predicting Variation in Emotional Valence and Arousal. We approached the task by fine-tuning a weighted ensemble of DeBERTa-v3-base models. Our system achieved the second-highest Valence composite correlation and ranked 5th in the overall V{\&}A average in Subtask 1. More importantly, we provide an empirical analysis of our model{'}s performance on longitudinal tasks, where it exhibited significant inverse cor- relations. We quantify the Venting Effect, showing a systematic tendency for the model to over-index on negative lexical cues despite self-reported relief. Furthermore, we analyze the structural trade-off between Mean Absolute Error and Pearson correlation induced by smoothing techniques."
}