Reproduce “Uncovering Cross-Domain Recommendation Ability of Large Language Models” 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

Xinyi Liu, Ruijie Wang, Dachun Sun, Dilek Hakkani-Tur, Tarek Abdelzaher

This paper from the University of Illinois Urbana-Champaign introduces LLM4CDR, a pipeline leveraging Large Language Models to perform cross-domain recommendations, specifically addressing extreme cold-start scenarios where no user purchase records are available in the target domain. It empirically demonstrates LLMs' ability to transfer knowledge across domains, with performance influenced by factors like history length, domain similarity, and prompt guidance.

Key insights

  • LLMs can transfer knowledge from source domains to address cold-start challenges in target domains for recommendation tasks.
  • The effectiveness of LLM-based cross-domain recommendation is influenced by factors such as the quantity of source history, candidate list size, and the conceptual gap between source and target domains.
  • Providing explicit domain-specific recommendation guidance and utilizing larger, more capable LLMs generally improves cross-domain recommendation performance.

Summary by alphaXiv.

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