Metraipolis › archive › dc_c729cb64ec2c

A Retrieval-Augmented Generation Method for Industrial Documents based on Reinforcement Learning (2026) · DOI 10.21203/rs.3.rs-9650882/v1

Source: Metraipolis archive
Original: https://doi.org/10.21203/rs.3.rs-9650882/v1
License (as archived): CC-BY-4.0
Archived: 2026-10-07T09:10:38.371Z
SHA-256 of the archived text: 0aadfb7fe5b08eeb89a4532d547f61505ffe797d6169749e644206d0c9ecccde
Stable URL of this page: https://metraipolis.com/documento/dc_c729cb64ec2c

How to cite

A Retrieval-Augmented Generation Method for Industrial Documents based on Reinforcement Learning (2026) · DOI 10.21203/rs.3.rs-9650882/v1 — Metraipolis archive, https://doi.org/10.21203/rs.3.rs-9650882/v1 (CC-BY-4.0); archived copy: https://metraipolis.com/documento/dc_c729cb64ec2c (Metraipolis archive dc_c729cb64ec2c, sha256 0aadfb7fe5b08eeb89a4532d547f61505ffe797d6169749e644206d0c9ecccde)

BibTeX

@misc{dcc729cb64ec2c,
  title = {A Retrieval-Augmented Generation Method for Industrial Documents based on Reinforcement Learning (2026) · DOI 10.21203/rs.3.rs-9650882/v1},
  howpublished = {Metraipolis archive},
  url = {https://doi.org/10.21203/rs.3.rs-9650882/v1},
  note = {License CC-BY-4.0; archived copy https://metraipolis.com/documento/dc_c729cb64ec2c; sha256 0aadfb7fe5b08eeb89a4532d547f61505ffe797d6169749e644206d0c9ecccde},
  year = {2026}
}

Archived text

Reproduced under CC-BY-4.0. Attribution: Metraipolis archive, https://doi.org/10.21203/rs.3.rs-9650882/v1. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Articolo recente con licenza aperta: scheda dai metadati Crossref (posted-content).
Titolo: A Retrieval-Augmented Generation Method for Industrial Documents based on Reinforcement Learning
Autori: Ling WeiQing, Yongqi Zhi
Pubblicato: 2026-09-15 · editore: Springer Science and Business Media LLC
DOI: https://doi.org/10.21203/rs.3.rs-9650882/v1
Licenza dell'articolo: CC-BY-4.0

Abstract (testo degli autori, CC-BY-4.0):
Abstract To address the challenges of semantic fragmentation and incomplete knowledge retrieval caused by information fragmentation in existing Large Language Model (LLM)-based Retrieval-Augmented Generation (RAG) methods, this paper proposes a Reinforcement Learning with LoRA-enhanced Multi-Round Retrieval Augmented Generation (RLo-RAG) model. The method first takes user queries and historical retrieval information as state inputs, employs a reinforcement learning-trained query generator to iteratively retrieve high-relevance document chunks, and generates answers using an LLM. A dynamic reward function is introduced during retrieval, combining similarity scoring, redundancy penalties, and efficiency rewards with adaptive weights to balance retrieval quality and efficiency. Evaluations on the HotPotQA and HoVer datasets demonstrate that RLo-RAG outperforms traditional RAG methods and multi-hop reasoning models. Case studies conducted via a prototype system further validate that the proposed approach improves retrieval precision and generation quality for industrial document, enhancing both accuracy and semantic integrity.

Fonte: https://api.crossref.org/works/10.21203%2Frs.3.rs-9650882%2Fv1 — metadati Crossref, consultati il 2026-10-06. Interrogazione: «retrieval augmented generation».