Reproduce “MeLU: Meta-Learned User Preference Estimator for Cold-Start 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
Hoyeop Lee, Jinbae Im, Seongwon Jang, Hyunsouk Cho, Sehee Chung
MeLU, developed by NCSOFT's Knowledge AI Lab, applies meta-learning (MAML) to address the user cold-start problem in recommender systems. It enables rapid personalization with minimal initial user data and features a strategy for intelligently selecting distinguishing items for preference elicitation, showing an MAE reduction of at least 5.92% and improved user satisfaction in subsequent recommendations.
Key insights
- Meta-learning, particularly MAML, can effectively adapt a preference model to new users with only a few observed interactions, addressing the user cold-start problem.
- A strategy combining gradient information and item popularity can intelligently select initial 'evidence candidates' that efficiently elicit a user's preferences, even if less broadly familiar.
- Personalized preference models can be generated directly for each user based on their unique, limited consumption history, rather than relying on similar users.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.