Reproduce “Leveraging Large Language Models in Conversational Recommender Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Luke Friedman, Sameer Ahuja, David Allen, Zhenning Tan, Hakim Sidahmed, Changbo Long, Jun Xie, Gabriel Schubiner and 5 more

RecLLM is an architecture for conversational recommendation that combines a large language model with external retrieval, ranking, explanation, and persistent natural-language user profiles. A LaMDA-based YouTube prototype demonstrates multi-turn preference refinement, topic shifts, profile-aware personalization, and recommendation explanations, while the paper outlines synthetic-user training methods for systems developed without production interaction logs.

Key insights

  • LLMs are most practical in large-scale recommendation when used for language understanding and interaction while external systems handle catalog retrieval and grounding.
  • Natural-language memory provides an editable alternative to opaque user embeddings and can support context-sensitive personalization, including overriding outdated preferences.
  • Synthetic users can help bootstrap a conversational recommender before production logs exist, but their realism, diversity, bias, and alignment with real user behavior require careful validation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.