Reproduce “Profiling What Matters: Context-Aware Item Profiles from Large-Scale Metadata for LLM Recommenders” 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
Dojun Hwang, Seunghan Lee, Cheonyoung Park, Sara Yu, SeongKu Kang
Researchers from Korea University and KT Corporation developed CAIRO (Context-Aware Item pROfiling), a framework for generating user-contextualized and structured item profiles from extensive metadata to enhance Large Language Model (LLM) recommenders. CAIRO consistently improved nDCG@5 by up to 12.84% over existing LLM baselines on Amazon datasets while maintaining practical online serving latency.
Key insights
- Simply injecting all available item features into LLMs can degrade reranking performance, underscoring the necessity for structured, selective, and context-aware item profiling.
- User-adaptive item profiles, which dynamically select features and traits based on individual user context, are crucial for enhancing LLM-based reranking accuracy.
- Pre-processing extensive item metadata offline into structured, granular components allows for efficient and low-latency online context-aware profiling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.