@inproceedings{hu-2026-kirito,
    title = "kirito at {S}em{E}val-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis via Sentence Structure Parsing Preprocessing and Prompt-Enhanced Instruction Tuning",
    author = "Hu, Shuangjin",
    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-1.125/",
    pages = "913--918",
    ISBN = "979-8-89176-414-9",
    abstract = "Dimensional Aspect-Based Sentiment Analysis (DimABSA) integrates fine-grained aspect extraction with continuous Valence{--}Arousal (VA) regression, posing unique challenges for fine-grained opinion mining. This paper presents our system for SemEval-2026 Task 3, with task-aligned strategies for three heterogeneous subtasks. For the DimASR task, we frame dimensional sentiment prediction as a supervised regression problem, paired with Low-Rank Adaptation (LoRA)-based parameter-efficient fine-tuning and a deep nonlinear regression head. For DimASTE and DimASQP tasks, we propose a lightweight sentence structure parsing preprocessing module, combined with prompt-enhanced instruction tuning for unified structured generation of aspect elements and VA scores. Experimental results on the official English test sets show that our system outperforms both official baselines across most settings, with syntax-guided prompting effectively improving aspect-opinion alignment and the dedicated regression head enhancing continuous sentiment modeling stability."
}