Reproduce “Coarse-to-fine Dynamic Uplift Modeling for Real-time Video 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
Chang Meng, Chenhao Zhai, Xueliang Wang, Shuchang Liu, Xiaoqiang Feng, Lantao Hu, Xiu Li, Han Li and 1 more
Kuaishou Technology and Tsinghua University developed a Coarse-to-fine Dynamic Uplift Modeling (CDUM) framework for real-time video recommendation, addressing the limitations of non-personalized modules by integrating both long-term user preferences and real-time interests. The deployed model on Kuaishou's platform improved long-term user retention metrics by up to 0.048% and app usage time by 0.135%.
Key insights
- Effectively combining coarse-grained long-term user preferences with fine-grained real-time interests significantly enhances the adaptability and performance of recommendation systems.
- Strategic treatment design, such as adjusting video duration distributions, enables the application of multi-treatment uplift modeling in complex content recommendation scenarios.
- Specialized embeddings and multi-expert networks within the CPM module can improve generalization and better utilize treatment information in uplift modeling tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.