MemRec: Collaborative Memory for LLM-Based Recommender Systems
A new framework called MemRec introduces collaborative memory to LLM-based agentic recommender systems, addressing the limitation of isolated semantic memories. By connecting user-item co-engagements and peer relationships across the community, MemRec enables sharing of relational insights to uncover hidden preferences, especially for data-sparse users. The framework decouples memory management from reasoning to avoid context overload and noise. Published on arXiv (2601.08816), the paper proposes a paradigm shift from traditional collaborative filtering to agentic systems enhanced with collaborative memory.
Key facts
- MemRec is a framework for collaborative memory-augmented agentic recommender systems.
- It addresses the isolation of semantic memories in existing LLM-based agents.
- Collaborative memory connects user-item co-engagements and peer relationships.
- The approach aims to improve predictions for data-sparse users.
- MemRec decouples memory management from reasoning to reduce context overload.
- The paper is published on arXiv with identifier 2601.08816.
- It shifts from traditional collaborative filtering to LLM-based agentic systems.
- The framework introduces a new paradigm for sharing relational insights.
Entities
Institutions
- arXiv