Metraipolis › archive › dc_6673100bbea1
Learning to Retrieve Missing Evidence for Long-Term Memory QA — arXiv:2609.37443
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.37443
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T16:27:16.159Z
- SHA-256 of the archived text
55beaddb3e5c547fbfba6954204d7c749893278027531d080f4602feb99db7c0- Stable URL of this page
https://metraipolis.com/documento/dc_6673100bbea1
How to cite
Learning to Retrieve Missing Evidence for Long-Term Memory QA — arXiv:2609.37443 — arXiv (en), https://arxiv.org/abs/2609.37443 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6673100bbea1 (Metraipolis archive dc_6673100bbea1, sha256 55beaddb3e5c547fbfba6954204d7c749893278027531d080f4602feb99db7c0)
BibTeX
@misc{dc6673100bbea1,
title = {Learning to Retrieve Missing Evidence for Long-Term Memory QA — arXiv:2609.37443},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.37443},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6673100bbea1; sha256 55beaddb3e5c547fbfba6954204d7c749893278027531d080f4602feb99db7c0},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.37443. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Long-term memory enables language models to use past interactions in future conversations. However, evidence needed to answer a question may be scattered across distant turns, while the question itself omits clues needed to locate it. Retrieved facts can reveal these clues, motivating retrieval decisions conditioned on evidence already found. We introduce MERA (Missing-Evidence Retrieval Augmentation), which separates globally searchable memory from a question-specific evidence state. Verified evidence guides subsequent retrieval without restricting access to the global memory. We train a lightweight planner through reinforcement learning, rewarding queries that recover previously missing evidence. MERA achieves strong answer accuracy across Qwen3-30B and GPT-4o-mini backbones. With Qwen3-30B for evidence processing and answer generation, the trained 0.6B planner achieves 77.40% accuracy on LoCoMo and 71.29% on LongMemEval-S, exceeding a 30B planner without retrieval-grounded training by 4.10% and 3.96%, respectively. On LoCoMo, later retrieval rounds increase cumulative evidence recall from 55.5% to 80.5%. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.37443 — 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 2026-10-07T15:54:53.982Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).