Reproduce “Online Item Cold-Start Recommendation with Popularity-Aware Meta-Learning” 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

Yunze Luo, Yuezihan Jiang, Yinjie Jiang, Gaode Chen, Jingchi Wang, Kaigui Bian, Peiyi Li, Qi Zhang

Researchers from Peking University and Kuaishou Technology developed Popularity-Aware Meta-learning (PAM), an online framework that uses fixed popularity-based task segmentation and a cold-start enhancer to address the item cold-start problem in streaming recommender systems. Online A/B tests on a commercial platform showed PAM improved item appearance rate by 41.39% and Like-Through Rate by 60.45%.

Key insights

  • Fixed popularity-based task segmentation is crucial for making meta-learning computationally feasible for online streaming recommendations, bypassing the need for per-item fine-tuning.
  • The 'Cold-Start Task Enhancer' effectively addresses data sparsity for new items by simulating cold-start interactions from popular items and leveraging self-supervised learning.
  • For cold-start items, the model prioritizes content-based features, while for popular items, it effectively utilizes both rich behavior-based and content-based features.

Summary by alphaXiv.

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