Constructing Knowledge Graphs from Text Using Large Language Models: Scoping Review
SBC Reviews on Computer Science · 2026
DOI: 10.5753/reviews.2026.6738 ↗Kurz gefasst
Scoping Review in der Zeitschrift SBC Reviews on Computer Science, 2026. Verfasst von Giovanna Borges Bottino und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Constructing Knowledge Graphs from unstructured textual sources poses significant challenges due to the inherent ambiguity of natural language and the high cost and limited scalability of manual knowledge modeling. Recently, Large Language Models (LLMs) have emerged as a promising alternative for automating knowledge extraction and structuring, resulting in a rapidly expanding body of research. This article aims to provide a scoping review of methods using LLMs to construct Knowledge Graphs from text, map existing approaches, identify methodological patterns, and analyze their strengths and limitations, thereby synthesizing the state of the art. The review employs a systematic protocol consistent with PRISMA guidelines and examines 126 primary studies. The literature is categorized into four methodological groups: Ontology-Based, Prompt-Based, RAG-Based, and Hybrid Pipelines. Each category is analyzed with respect to the role of LLMs within the construction pipeline, the degree of semantic formalization, and the architectural strategies implemented. Comparative analysis is performed using five evaluation criteria: exactness, scalability, adaptability, reproducibility, and ease of implementation. The findings demonstrate that no single approach comprehensively addresses all challenges inherent to LLM-based Knowledge Graph construction. Ontology-Based methods provide robust semantic guarantees but demand considerable manual effort. Prompt-Based approaches facilitate rapid deployment yet exhibit variability and limited reproducibility. RAG-Based methods enhance grounding through external evidence but introduce reliance on retrieval mechanisms. Hybrid Pipelines yield higher extraction accuracy, though at the expense of increased architectural complexity. The review identifies substantial heterogeneity in evaluation practices and a lack of standardized quantitative metrics. These results underscore the necessity for consistent evaluation frameworks to enhance comparability, reliability, and the advancement of research in LLM-assisted Knowledge Graph construction.
Zitieren
APA
Bottino, G. B., & Alcazár, J. D. J. P. (2026). Constructing Knowledge Graphs from Text Using Large Language Models: Scoping Review. SBC Reviews on Computer Science. https://doi.org/10.5753/reviews.2026.6738
Vancouver
Bottino GB, Alcazár JDJP. Constructing Knowledge Graphs from Text Using Large Language Models: Scoping Review. SBC Reviews on Computer Science. 2026. doi:10.5753/reviews.2026.6738
BibTeX
@article{Bottino2026,
author = {Bottino, Giovanna Borges and Alcazár, José de Jesus Pérez},
title = {Constructing Knowledge Graphs from Text Using Large Language Models: Scoping Review},
journal = {SBC Reviews on Computer Science},
year = {2026},
doi = {10.5753/reviews.2026.6738}
}
EUVIMED