EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Granulated Integer Power-Based Contrastive Learning for Multilingual Stance Detection

Muhammad Ebrahim Ahmadi, Monireh Hosseini

Algorithms · 14.11.2026 · 19 (11) · 1004

DOI: 10.3390/a19111004 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Algorithms, online erschienen am 14.11.2026. Verfasst von Muhammad Ebrahim Ahmadi und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multilingual stance detection seeks to determine whether a text supports or opposes a given claim across different languages. The task remains challenging because languages differ in their linguistic structures, writing styles, and ways of expressing opinions, making it difficult to learn representations that generalize effectively across languages. This paper proposes a novel contrastive learning framework for multilingual stance detection. The proposed framework combines two complementary components. The first is a granulation algorithm based on the principle of minimum integer power, which generates multiple textual granules to enrich the feature space with diverse semantic patterns. This enables the model to generate informative training instances, particularly when the available data are limited. The second component is a supervised contrastive learning objective specifically designed for multilingual stance detection. While the granulation process increases feature diversity by generating semantically varied representations, the contrastive objective encourages the model to bring representations associated with similar stances closer together while separating those associated with different stances. These two components are integrated with a multilingual pre-trained language model to learn robust cross-lingual representations. Experimental results demonstrate that the proposed framework consistently outperforms existing approaches, achieving an accuracy of 72.52%, compared with the 53–61% accuracy achieved by conventional machine learning methods, deep learning models, and existing multilingual pre-trained models.

Zitieren

APA

Ahmadi, M. E., & Hosseini, M. (2026). Granulated Integer Power-Based Contrastive Learning for Multilingual Stance Detection. Algorithms, 19(11), 1004. https://doi.org/10.3390/a19111004

Vancouver

Ahmadi ME, Hosseini M. Granulated Integer Power-Based Contrastive Learning for Multilingual Stance Detection. Algorithms. 2026;19(11):1004. doi:10.3390/a19111004

BibTeX
@article{Ahmadi2026,
  author = {Ahmadi, Muhammad Ebrahim and Hosseini, Monireh},
  title = {Granulated Integer Power-Based Contrastive Learning for Multilingual Stance Detection},
  journal = {Algorithms},
  year = {2026},
  volume = {19},
  number = {11},
  pages = {1004},
  doi = {10.3390/a19111004}
}
Kostenlos starten →