Reproduce “KLAN: Kuaishou Landing-page Adaptive Navigator” 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
Fan Li, Chang Meng, Jiaqi Fu, Shuchang Liu, Jiashuo Zhang, Tianke Zhang, Xueliang Wang, Xiaoqiang Feng
Kuaishou Technology developed KLAN, a hierarchical system that adaptively combines long-term user preferences and short-term interests to personalize landing page assignments for users upon app entry. Deployed on the Kuaishou platform, KLAN reduced the Page Drop-off Ratio by 28.158% and increased Daily Active Users by 0.205% and Watch Time by 0.443%.
Key insights
- The use of treatment-specific branches with KL divergence regularization in KLAN-ISP effectively prevents gradient conflicts in multi-treatment causal uplift modeling, improving causal effect estimation for diverse landing pages.
- A Dynamic Conservative Coefficient in KLAN-IIT adaptively adjusts reinforcement learning policy conservatism based on real-time data confidence, enhancing stability for modeling volatile intra-day user interests.
- Adaptive modulation dynamically weighs inter-day static preferences and intra-day dynamic interests based on current contextual features, allowing for more flexible and personalized landing page assignments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.