Reproduce “Modeling Long-term User Behaviors with Diffusion-driven Multi-interest Network 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
Weijiang Lai, Beihong Jin, Yapeng Zhang, Yiyuan Zheng, Rui Zhao, Jian Dong, Jun Lei, Xingxing Wang
A novel architecture, DiffuMIN, leverages diffusion models to augment and explore user interests for Click-Through Rate (CTR) prediction, effectively modeling long-term user behaviors. It demonstrated a 1.52% increase in CTR and 1.10% increase in CPM in an online A/B test on Meituan's industrial platform.
Key insights
- Diffusion models can be effectively adapted to augment and thoroughly explore the latent space of user interests in CTR prediction, moving beyond conventional filtering methods.
- Orthogonal decomposition and target-oriented behavior routing are crucial for extracting distinct and diverse multi-interests from noisy long-term behavior sequences, minimizing inter-channel redundancy.
- Guiding the diffusion process with both contextual user interests and specific interest channels, and initiating generation from perturbed interests, significantly enhances the personalization and quality of generated augmented interests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.