EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

RAG-Enhanced DistilBERT Model for Contextual Understanding of Polysemy in Digital News Titles

Anusha, K. Manjula Shenoy, Smitha N. Pai

IEEE Access · 2026 · 14 · 90425-90443

DOI: 10.1109/access.2026.3702098 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Anusha und 2 weiteren Autor:innen. Die Arbeit nennt 75 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, India.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Digital news titles often include polysemous phrases to enhance linguistic efficiency. It offers major disambiguation challenges for computational natural language processing systems in media analysis and information retrieval applications. Large language models (LLMs) are better at reasoning in context, but they struggle with polysemy in news titles. These limitations are connected to the knowledge requirements that fail to account for the ongoing evolution of news-reporting language, as well as an inability to correct errors when headline styles are inconsistent. Also, constraints in contextual understanding of present developments and a greater reliance on language. Computing devices have limited efficiency in real-time news processing. This study introduces a retrieval-augmented generation method to improve the management of polysemous terms in digital news. The suggested method adeptly handles the challenges identified by LLMs through the use of external information sources, journalism databases, and contextual knowledge repositories. Understanding massive language models for digital news by using adaptive retrieval techniques enables accurate disambiguation of polysemous titles through integrated knowledge processes. The suggested research indicates that retrieval-augmented generation (RAG)-enhanced frameworks can effectively address the challenges faced by LLMs in news-related tasks, demonstrating their performance in automated journalism analysis, headline-generating systems, and news categorization and summarization applications. The proposed RAG-DistilBERT model achieves 11.7% improvement in accuracy over standard models for the contextual interpretation of digital news titles.

Förderung

  • Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, India

Zitieren

APA

Anusha, Shenoy, K. M., & Pai, S. N. (2026). RAG-Enhanced DistilBERT Model for Contextual Understanding of Polysemy in Digital News Titles. IEEE Access, 14, 90425-90443. https://doi.org/10.1109/access.2026.3702098

Vancouver

Anusha, Shenoy KM, Pai SN. RAG-Enhanced DistilBERT Model for Contextual Understanding of Polysemy in Digital News Titles. IEEE Access. 2026;14:90425-90443. doi:10.1109/access.2026.3702098

BibTeX
@article{Anusha2026,
  author = {Anusha and Shenoy, K. Manjula and Pai, Smitha N.},
  title = {RAG-Enhanced DistilBERT Model for Contextual Understanding of Polysemy in Digital News Titles},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {90425--90443},
  doi = {10.1109/access.2026.3702098}
}
Kostenlos starten →