Reproduce “Review-driven Personalized Preference Reasoning with Large Language Models for Recommendation” 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

Jieyong Kim, Hyunseo Kim, Hyunjin Cho, SeongKu Kang, Buru Chang, Jinyoung Yeo, Dongha Lee

Researchers from Yonsei and Korea Universities developed Exp3rt, an LLM-based recommender system that creates structured user and item preference profiles from reviews to enhance rating prediction and generate logical explanations. Leveraging knowledge distillation from GPT-3.5 to LLaMA3-8B, Exp3rt consistently outperformed baselines in rating prediction and top-k recommendation while producing significantly more persuasive and faithful explanations.

Key insights

  • Integrating rich, structured preference profiles derived from user and item reviews enables LLMs to develop a more nuanced and comprehensive understanding of preferences compared to sparse ID-based or simple textual inputs.
  • Implementing explicit step-by-step textual reasoning within LLMs is crucial for both achieving higher accuracy in rating prediction and generating faithful, logical, and persuasive recommendation explanations.
  • Knowledge distillation effectively transfers complex reasoning capabilities from large, powerful teacher LLMs to smaller, more computationally efficient student models, making advanced LLM-based recommenders practical for domain-specific tasks.

Summary by alphaXiv.

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