EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

EMEF: An evidence-grounded multi-view extraction framework for fine-grained agricultural knowledge organization

Huan Yuan, Zhaoye Zhou, Chenxue Yang, Beilei Fan, Qingbo Zhou, Zhemin Li, Xian Li

Smart Agricultural Technology · 2026-12 · 15 · 102565

DOI: 10.1016/j.atech.2026.102565 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Huan Yuan und 6 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: Central Public-interest Scientific Institution Basal Research Fund, Chinese Academy of Fishery Sciences, National Key Research and Development Program of China, Key Technologies Research and Development Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Agricultural scientific literature encompasses abundant multi-category knowledge supporting agricultural production management. Nevertheless, conventional knowledge graphs are built upon static entity-relation triples and overlook deep-level knowledge including events, temporal sequences, and causality relationships. This study proposed an evidence-grounded multi-view extraction framework (EMEF) for fine-grained agricultural knowledge organization. EMEF enables zero-shot extract static, event, temporal and causal knowledge synchronously. It further integrates paragraph-level evidence anchoring, a repair-candidate strategy and non-destructive global soft alignment to improve knowledge attribute discrimination, evidence credibility and context retention. Compared with extraction limited solely to static triples, EMEF yields a 47.39% increase in the number of knowledge units. On an evaluation dataset consisting of 100 randomly sampled paragraphs, the repair-candidate strategy retained 76 valid knowledge units and 48 candidate knowledge units on contrast to the strict-drop strategy, achieving an entity compression rate of approximately 35.26%. When combined with EMEF, the macro-average F 1 score of the four knowledge types extracted by GPT-5.4 was improved by 67.54%. By establishing separate extraction and validation boundaries for the four categories of knowledge, EMEF outperforms of AutoSchemaKG, OneKE, and KG-Gen by approximately 51.9% in extraction accuracy. Finally, EMEF was deployed on 140,430 wheat-related literature, leading to a large-scale wheat knowledge graph comprising 23,957,944 nodes and 22,422,598 relationships. Its well-defined, high-granularity, and high-aggregation-density characteristics manifested that fragmented agricultural knowledge has been organically organized. As a zero-shot framework, EMEF is adaptable to diverse crop varieties and knowledge practice scenarios, paving the way for the vertical deployment of large language models.

Förderung

  • Central Public-interest Scientific Institution Basal Research Fund, Chinese Academy of Fishery Sciences
  • National Key Research and Development Program of China
  • Key Technologies Research and Development Program

Zitieren

APA

Yuan, H., Zhou, Z., Yang, C., Fan, B., Zhou, Q., Li, Z., & Li, X. (2026). EMEF: An evidence-grounded multi-view extraction framework for fine-grained agricultural knowledge organization. Smart Agricultural Technology, 15, 102565. https://doi.org/10.1016/j.atech.2026.102565

Vancouver

Yuan H, Zhou Z, Yang C, Fan B, Zhou Q, Li Z, et al. EMEF: An evidence-grounded multi-view extraction framework for fine-grained agricultural knowledge organization. Smart Agricultural Technology. 2026;15:102565. doi:10.1016/j.atech.2026.102565

BibTeX
@article{Yuan2026,
  author = {Yuan, Huan and Zhou, Zhaoye and Yang, Chenxue and Fan, Beilei and Zhou, Qingbo and Li, Zhemin and Li, Xian},
  title = {EMEF: An evidence-grounded multi-view extraction framework for fine-grained agricultural knowledge organization},
  journal = {Smart Agricultural Technology},
  year = {2026},
  volume = {15},
  pages = {102565},
  doi = {10.1016/j.atech.2026.102565}
}
Kostenlos starten →