rEDMRec: Distilling Large Language Model Reasoning into an Editable Experience Memory for Recommendation

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Minh Hoang Nguyen, Tung Le, Huy Tien Nguyen
• Venue: arXiv: cs.IR
• Published: 2026-08-19
• Collected: 2026-08-20
• URL: http://arxiv.org/abs/2608.18952v1
• PDF: https://arxiv.org/pdf/2608.18952v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The paper proposes a new approach for recommendation systems, called rEDMRec, which distills the reasoning of large language models into an editable memory that can be used to rank candidate items without regenerating the reasoning each time. The system is composed of four channels: long-term preference, short-term context, item-perception, and counterfactual hard-negative comparisons. It uses a lightweight student model and a more complex teacher model for learning and optimization. The authors claim that rEDMRec improves recommendation quality over several datasets and models, with improvement up to 13.3% compared to the second-best baseline on one dataset.

What the Document Says

• rEDMRec is a new approach for recommendation systems.
• It distills reasoning of large language models into four channels: long-term preference, short-term context, item-perception, and counterfactual hard-negative comparisons.
• An LLM memory controller manages the channels with Add/Delete/Modify/Keep operations and refines entries via K-agent debate.
• A lightweight student LLM ranks candidates purely by retrieving from this memory.
• rEDMRec improves HR@1 (Hit Ratio at rank 1) over several datasets and models, compared to zero-shot, few-shot, and RAG (Ranking with Auxiliary Gradients), and over GraphRAG on most backbones.
• Channel ablations show that short-term context is the only channel that helps consistently across capacity tiers.

Key Concepts Mentioned

• recommendation systems
• large language models
• rEDMRec
• LLM memory controller
• K-agent debate
• HR@1

Claims Made by the Document

• rEDMRec improves HR@1 over several datasets and models.
• Channel ablations show that short-term context is the only channel that helps consistently across capacity tiers.

Evidence Used

• Improvements up to 13.3% vs. the second-best baseline on ML-1M
• Results across ML-1M, Amazon Beauty, and Steam, and ten student backbones

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Large language models can improve recommendation quality by reasoning explicitly over user history and candidate items - for example, extracting a user's preferences or explaining why one item fits better than another - rather than mapping history directly to a ranked list. This reasoning, however, is expensive to repeat on every ranking request and, once produced, is typically consumed once and discarded, leaving it neither reusable across future requests nor easy to inspect or correct as user tastes drift. Our insight is that reasoning does not need to be regenerated at every call if it can instead be compressed once into a compact, structured memory that a lightweight model retrieves from. We propose rEDMRec, which distills a teacher LLM's reasoning into four typed, editable experience channels - long-term preference, short-term context, item-perception, and counterfactual hard-negative comparisons - maintained by an LLM memory controller that performs Add/Delete/Modify/Keep operations and refines entries via K-agent debate. A lightweight student LLM then ranks candidates purely by retrieving from this memory, without invoking the teacher again, decoupling online inference cost from reasoning depth. Across ML-1M, Amazon Beauty, and Steam and ten student backbones, rEDMRec improves HR@1 over zero-shot, few-shot, and RAG on every backbone, and over GraphRAG on most backbones, with Impv up to 13.3% vs. the second-best baseline on ML-1M. Channel ablations show that short-term context is the only channel that helps consistently across capacity tiers, whereas long-term, item-perception, and counterfactual contributions are capacity-dependent (and can reverse on the strongest students); debate-based memory optimization lowers bank duplication by 7.4 percentage points while raising downstream HR@1 by up to +0.029 over six optimization epochs.