Reproduce “POSO: Personalized Cold Start Modules for Large-scale Recommender Systems” 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
Shangfeng Dai, Haobin Lin, Zhichen Zhao, Jianying Lin, Honghuan Wu, Zhe Wang, Sen Yang, Ji Liu
The POSO framework introduces an architectural approach to reinforce personalization in large-scale recommender systems, preventing personalized features from being ignored for minority groups like new users and items. Deployed on Kwai, this method improved new user watch time by 7.75% and retention by 1.52%, alongside a 3.81% increase in new video watch time.
Key insights
- Personalized features for minority user groups can be "submerged" and ignored by deep learning models when training on heavily imbalanced datasets, leading to suboptimal recommendations.
- Explicitly reinforcing personalization via specialized sub-modules and dedicated gating networks can effectively model the distinct behaviors of new and inactive users without significant computational overhead.
- The selection of specific "Personalization Codes," such as bucketized accumulated view count, is vital for effectively driving the personalized gating mechanism and ensuring module specialization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.