@inproceedings{cao-etal-2026-hus,
    title = "{HUS}@{NLP}-{VNU} at {S}em{E}val-2026 Task 3: Dual-Stream Syntax-Aware Modeling and Direct Preference Optimization for Dimensional {ABSA}",
    author = "Cao, An  and
      Hoang, Lam  and
      Toan, Le Ngoc  and
      Linh, Ha",
    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-1.278/",
    pages = "2200--2208",
    ISBN = "979-8-89176-414-9",
    abstract = "Dimensional Aspect-Based Sentiment Analysis (DimABSA) extends traditional ABSA by predicting continuous sentiment intensity in the Valence-Arousal space. To tackle the regression subtasks (DimASR and DimStance), we propose a Dual-Stream Syntax-Aware architecture synergizing contextual semantics with a Deep Syntax-Guided Graph Convolutional Network (GCN). It utilizes a Context-Aware Anchor for semantic filtering and post-norm residuals to prevent oversmoothing. For generative extraction, we apply Direct Preference Optimization (DPO) via a resource-efficient, heuristic-based data perturbation strategy to construct preference pairs without costly LLMs. Across multilingual settings, our regression model achieves top-5 rankings in nine domains and obtains the best result on the Chinese-Finance dataset. Empirical analysis shows that explicit syntactic modeling consistently improves continuous sentiment regression, while DPO provides modest but stable gains for boundary-constrained extraction."
}