EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

DomainBench-LLM: Benchmarking Domain-Specific Language Models for Specialized NLP Tasks

Sofia Nystrom, Sofia Bellamy, Arthur Penhaligon

INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES · 10.04.2026 · 6 (1) · 47-56

DOI: 10.66238/ijrss.v6i1.91 ↗

Kurz gefasst

Fachartikel in der Zeitschrift INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES, online erschienen am 10.04.2026. Verfasst von Sofia Nystrom und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid proliferation of large language models has fundamentally transformed the landscape of natural language processing. While general-purpose models exhibit remarkable capabilities across a broad spectrum of everyday tasks, their efficacy in highly specialized domains such as medicine, law, finance, and advanced scientific research remains a subject of intense scrutiny. The primary challenge arises from the lexical, semantic, and structural idiosyncrasies inherent to specialized corpora, which often diverge significantly from the generic web-crawled text predominantly used during the pre-training phase. Consequently, evaluating the true utility of these models requires benchmarking frameworks that transcend basic reasoning and common-sense question answering. In this paper, we introduce DomainBench-LLM, a novel, comprehensive benchmarking suite specifically engineered to evaluate the performance, robustness, and computational efficiency of language models in specialized technical domains. Our framework incorporates a meticulously curated dataset encompassing thousands of specialized tasks, utilizing a multidimensional evaluation metric that accounts for factual accuracy, domain-specific semantic alignment, and inference efficiency. Through extensive empirical analysis of several state-of-the-art models, we demonstrate that while general models frequently hallucinate or misinterpret technical jargon, domain-adapted models achieve significantly higher precision, albeit with varying computational trade-offs. The results emphasize the critical necessity for specialized evaluation paradigms and provide a standardized methodology for future research in domain-specific natural language processing.

Zitieren

APA

Nystrom, S., Bellamy, S., & Penhaligon, A. (2026). DomainBench-LLM: Benchmarking Domain-Specific Language Models for Specialized NLP Tasks. INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES, 6(1), 47-56. https://doi.org/10.66238/ijrss.v6i1.91

Vancouver

Nystrom S, Bellamy S, Penhaligon A. DomainBench-LLM: Benchmarking Domain-Specific Language Models for Specialized NLP Tasks. INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES. 2026;6(1):47-56. doi:10.66238/ijrss.v6i1.91

BibTeX
@article{Nystrom2026,
  author = {Nystrom, Sofia and Bellamy, Sofia and Penhaligon, Arthur},
  title = {DomainBench-LLM: Benchmarking Domain-Specific Language Models for Specialized NLP Tasks},
  journal = {INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES},
  year = {2026},
  volume = {6},
  number = {1},
  pages = {47--56},
  doi = {10.66238/ijrss.v6i1.91}
}
Kostenlos starten →