@inproceedings{missaoui-etal-2026-dualaxis,
    title = "{D}ual{A}xis {AI} at {S}em{E}val-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis",
    author = "Missaoui, Yahya  and
      Kebede, Solomon  and
      Marreddy, Mounika  and
      Mehler, Alexander",
    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-1.82/",
    pages = "573--578",
    ISBN = "979-8-89176-414-9",
    abstract = "Dimensional Aspect-Based Sentiment Analy-sis models sentiment using continuous valenceand arousal scores instead of discrete polaritylabels, enabling fine-grained affect representa-tion at the aspect level. SemEval 2026 Task3 defines this setting through three subtaskscovering aspect-level regression and structuredextraction of aspect{--}opinion pairs with continu-ous scoring. We implement transformer-basedbaselines for all subtasks within a unified, re-producible framework. For aspect-level regres-sion, we fine-tune pretrained encoders in anaspect-conditioned setup to predict valence andarousal. RoBERTa-large achieves the best de-velopment performance, with average RMSEsof 0.884 (restaurant) and 0.789 (laptop)."
}