Metraipolis › archive › dc_f6d689f141da

PRO-Step: Step-level Process Reward Optimization for Retrieval-Augmented Generation — arXiv:2609.01658

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.01658
License (as archived)
CC0-1.0
Archived
2026-10-17T03:26:47.717Z
SHA-256 of the archived text
dcf3b6af1d40896d2dd802c4f4d2324efafc7407be82298f5e3a804f1c8098e2
Stable URL of this page
https://metraipolis.com/documento/dc_f6d689f141da

How to cite

PRO-Step: Step-level Process Reward Optimization for Retrieval-Augmented Generation — arXiv:2609.01658 — arXiv (en), https://arxiv.org/abs/2609.01658 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f6d689f141da (Metraipolis archive dc_f6d689f141da, sha256 dcf3b6af1d40896d2dd802c4f4d2324efafc7407be82298f5e3a804f1c8098e2)
BibTeX
@misc{dcf6d689f141da,
  title = {PRO-Step: Step-level Process Reward Optimization for Retrieval-Augmented Generation — arXiv:2609.01658},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.01658},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f6d689f141da; sha256 dcf3b6af1d40896d2dd802c4f4d2324efafc7407be82298f5e3a804f1c8098e2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01658. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Retrieval-Augmented Generation enhances Large Language Models by grounding responses in external knowledge, but multi-hop reasoning remains vulnerable to error propagation, where early retrieval failures confound subsequent steps. Standard outcome-based optimization only rewards the final answer, leaving intermediate retrieval and reasoning errors undetected. While existing process-based methods introduce step-level signals, they still score each step against the final answer, rewarding spurious successes where flawed retrieval coincidentally produces the correct answer. Step-level supervision in RAG requires evaluating both logical validity and evidential grounding at each step. We introduce PRO-STEP: we train a generative PRM that evaluates both dimensions, employ PRM-guided value tree search to construct preference pairs contrasting valid steps against flawed ones, and optimize the policy via step-level Direct Preference Optimization. Experiments on single and multi-hop QA datasets demonstrate that PRO-STEP achieves the best average EM and F1 across five benchmarks. Code, models, and training data are publicly available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-03 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.01658 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T03:13:12.467Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-03?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).