Reproduce “Reweighting Clicks with Dwell Time in 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
Ruobing Xie, Lin Ma, Shaoliang Zhang, Feng Xia, Leyu Lin
WeChat, Tencent researchers developed a click reweighting framework for recommendation systems, utilizing a novel 'valid read' definition and a 'normalized dwell time function'. This approach, deployed on WeChat Top Stories, resulted in a 2.577% increase in Click-Through Rate (CTR) and a 4.139% increase in average clicks, demonstrating improved quality and efficiency in user engagement.
Key insights
- Defining 'valid read' adaptively provides a more robust and high-quality signal of user engagement compared to treating all clicks equally, effectively filtering out misleading interactions.
- Normalizing dwell time with a sigmoid function allows for a nuanced quantification of user satisfaction, prioritizing efficient, meaningful engagement over simply maximizing time spent.
- Reweighting click signals directly within an MTL framework based on these refined definitions can guide recommendation models towards generating more engaging and satisfying content for users.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.