Reproduce “An Efficient Embedding Based Ad Retrieval with GPU-Powered Feature Interaction” 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
Yifan Lei, Jiahua Luo, Tingyu Jiang, Bo Zhang, Lifeng Wang, Dapeng Liu, Zhaoren Wu, Haijie Gu and 2 more
Researchers at Tencent Inc. developed an ad retrieval system that integrates implicit and explicit feature interactions into dual-tower networks. This system uses a novel GPU-accelerated inverted list for explicit interactions, achieving a 2.62% GAUC improvement offline and up to 1.58% GMV increase online while maintaining sub-millisecond latency.
Key insights
- Combining both implicit (IPNN) and explicit (HitMatch) feature interactions significantly enhances retrieval accuracy compared to baseline dual-tower models.
- Explicit cross-features, particularly from side information and historical user interactions, provide strong signals for improving ad relevance during retrieval.
- Specialized GPU-accelerated data structures can overcome the computational barriers of complex interaction models, making them viable for large-scale, real-time ad retrieval.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.