Reproduce “Generative Sequential Recommendation via Hierarchical Behavior Modeling” 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
Zhefan Wang, Guokai Yan, Jinbei Yu, Siyu Gu, Jingyan Chen, Peng Jiang, Zhiqiang Guo, Min Zhang
Researchers from Tsinghua University and Kuaishou Technology developed GAMER, a generative framework for multi-behavior sequential recommendation that explicitly models hierarchical dependencies among user behaviors and employs a sequential augmentation strategy. Evaluated on a new large-scale dataset, ShortVideoAD, GAMER achieved significant performance gains, including a +2.5% increase in Conversion Rate and +1.8% in effective Cost Per Mille during online A/B tests on a short-video platform.
Key insights
- Explicitly modeling the hierarchical dependencies between different user behavior types within a generative framework significantly improves recommendation performance for high-value conversions.
- A targeted sequential augmentation strategy, which selectively removes lower-level behaviors, enhances model robustness and generalization, especially for sparse high-level behaviors.
- Decoder-only generative architectures, combined with user-level training granularity, offer substantial training efficiency gains compared to encoder-decoder models in multi-behavior settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.