Reproduce “Deep Group Interest Modeling of Full Lifelong User Behaviors for CTR Prediction” 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
Qi Liu, Xuyang Hou, Haoran Jin, Xiaolong Chen, Jin Chen, Defu Lian, Zhe Wang, Jia Cheng and 1 more
Researchers from USTC, UESTC, and Meituan developed the Deep Group Interest Network (DGIN), an end-to-end model that efficiently processes a user's full lifelong, multi-type behaviors for Click-Through Rate prediction without discarding information. Deployed in Meituan's LBS advertising system, DGIN led to a 4.5% increase in CTR and a 2.0% improvement in Revenue per Mille, with minimal increases in latency and storage.
Key insights
- Efficient grouping of behaviors by 'interest key' enables end-to-end training on full lifelong sequences without explicit retrieval, ensuring all behavior information contributes to interest learning.
- Enriching group representations with both statistical (intensity) and aggregated (evolutionary) attributes is crucial for a comprehensive and unbiased understanding of long-term user preferences.
- Separately modeling general long-term interest (GM) and candidate-specific decision interest (TM) provides a more nuanced and accurate prediction of user engagement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.