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 authors propose a system called rEDMRec that improves recommendation quality by compressing the reasoning of a large language model (LLM) into a compact, editable memory. This memory is maintained by an LLM memory controller and can be accessed by a lightweight student LLM for candidate ranking. The system shows improvements in various datasets and models, with channel ablations indicating that short-term context is consistently beneficial across capacity tiers.

What the Document Says

• The authors propose rEDMRec, which distills reasoning from a teacher LLM into four editable experience channels.
• These channels are long-term preference, short-term context, item-perception, and counterfactual hard-negative comparisons.
• An LLM memory controller maintains these channels and performs Add/Delete/Modify/Keep operations.
• A lightweight student LLM ranks candidates by retrieving from this memory.
• rEDMRec shows improvements in HR@1 over various baselines on multiple datasets and models.
• Channel ablations show that short-term context is beneficial across capacity tiers, while long-term, item-perception, and counterfactual contributions are capacity-dependent.
• Debate-based memory optimization lowers bank duplication and raises downstream HR@1.

Key Concepts Mentioned

• Large Language Models
• Recommendation System
• rEDMRec
• Experience Channels
• LLM Memory Controller
• Student LLM
• HR@1
• Channel Ablations
• Debate-based Memory Optimization

Claims Made by the Document

• rEDMRec improves HR@1 over various baselines.
• Short-term context is beneficial across capacity tiers.
• Long-term, item-perception, and counterfactual contributions are capacity-dependent.

Evidence Used

• Improvements 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.

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.