@inproceedings{laschenko-korotyk-2026-sokratum,
    title = "{S}okra{TUM} at {S}em{E}val-2026 Task 3: A hybrid cascade of Label Distribution Learning, {RAG} supported generative extraction and contrastive metric learning for dimensional sentiment analysis",
    author = "Laschenko, Denis  and
      Korotyk, Albert",
    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-1.241/",
    pages = "1919--1929",
    ISBN = "979-8-89176-414-9",
    abstract = "The Dimensional ABSA (DimABSA) sharedtask extends traditional aspect-based sentimentanalysis from categorical polarity to continuousvalence{--}arousal (VA) prediction. We presentour system for all three subtasks: DimensionalAspect Sentiment Regression (DimASR),Dimensional Aspect Sentiment Triplet Extrac-tion (DimASTE), and Dimensional AspectSentiment Quad Prediction (DimASQP).Due to the cascading nature of the differentsubtasks, we built a modular interlockingpipeline that uses classical Machine Learningand NLP methods.Experiments across domains show consistentgains in regression accuracy and structuredextraction performance. Our results demon-strate the effectiveness of distribution-awareregression, retrieval-augmented generation, andcontrastive prototype learning for dimensionalsentiment analysis."
}