@inproceedings{affan-etal-2026-habib,
    title = "Habib University at {S}em{E}val-2026 Task 3: A Pipeline Approach for Dimensional Aspect-Based Sentiment Analysis",
    author = "Affan, Muhammad  and
      Shahzad, M Hassan  and
      Imam, Mikaal  and
      Zulfiqar, Moiz  and
      Kumar, Sandesh  and
      Samad, Abdul",
    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-1.428/",
    pages = "3449--3459",
    ISBN = "979-8-89176-414-9",
    abstract = "Aspect-based sentiment analysis has evolved from categorical polarity classification to fine-grained modeling of continuous affective dimensions. Dimensional Aspect-Based Sentiment Analysis (DimABSA) extends this paradigm by requiring both structured sentiment extraction and continuous valence{--}arousal (VA) regression in multilingual settings. In this paper, we present our system for SemEval-2026 Task 3, which evaluates this challenge across six languages and four domains, requiring systems to extract aspect{--}category{--}opinion quadruplets and predict VA scores on a 1{--}9 scale.We propose a modular four-stage multilingual transformer pipeline for element extraction, aspect{--}opinion pairing, category prediction, and VA regression. We conduct experiments over multiple models and training configurations, including VA rescaling to [-1,1], Gaussian label noise injection, Concordance Correlation Coefficient (CCC) loss, and Savitzky{--}Golay smoothing. Among all languages, our system achieves the lowest RMSE of 0.5333 on Subtask 1 and the highest cF1 of 0.5492 on Subtask 2. We further investigate data augmentation to improve low-resource performance and address label imbalance. Ultimately, our modular architecture demonstrated highly competitive cross-lingual transfer, achieving top-tier placements in low-resource settings, including 2nd place for Tatar and 6th place for Russian in dimensional regression."
}