Reference-Grounded User Feedback and Guarded Memory Updating for Agricultural Knowledge Services: A Controlled Study with an Exploratory Local-Model Stress Test
Applied Sciences · 20.11.2026 · 16 (22) · 11518
DOI: 10.3390/app162211518 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 20.11.2026. Verfasst von Yi Wang und 4 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Ministry of Agriculture and Rural Affairs.
Abstract (Originalsprache)
User feedback can help agricultural question-answering services adapt. Storing an unverified correction can also carry an error into later answers. We examined these two uses of feedback separately using 160 judgments from eight participants, controlled simulations, and exploratory language-model tests. In simulation, the full router combined reliability weighting with replay and showed no statistically distinguishable advantage over equal weighting. In the local-model test, unfiltered corrections changed 16 of 78 responses from matching the original key to disagreeing with it. Alternative reference judgments changed the flip counts, and the server-model results also depended on how agreement was scored. Workflow tests blocked stale updates and restored an earlier snapshot, while erroneous approvals could still enter memory. The results support separating feedback used for response selection from the verification required before a correction is published. A performance advantage for the full framework and safety in agricultural practice remain unestablished.
Förderung
- Ministry of Agriculture and Rural Affairs
Zitieren
APA
Wang, Y., Zhao, X., Xiao, Y., Hao, P., & Wu, H. (2026). Reference-Grounded User Feedback and Guarded Memory Updating for Agricultural Knowledge Services: A Controlled Study with an Exploratory Local-Model Stress Test. Applied Sciences, 16(22), 11518. https://doi.org/10.3390/app162211518
Vancouver
Wang Y, Zhao X, Xiao Y, Hao P, Wu H. Reference-Grounded User Feedback and Guarded Memory Updating for Agricultural Knowledge Services: A Controlled Study with an Exploratory Local-Model Stress Test. Applied Sciences. 2026;16(22):11518. doi:10.3390/app162211518
BibTeX
@article{Wang2026,
author = {Wang, Yi and Zhao, Xiaoheng and Xiao, Yuqi and Hao, Peng and Wu, Huarui},
title = {Reference-Grounded User Feedback and Guarded Memory Updating for Agricultural Knowledge Services: A Controlled Study with an Exploratory Local-Model Stress Test},
journal = {Applied Sciences},
year = {2026},
volume = {16},
number = {22},
pages = {11518},
doi = {10.3390/app162211518}
}
EUVIMED