Reproduce “Should I send this notification? Optimizing push notifications decision making by modeling the future” 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

Conor O'Brien, Huasen Wu, Shaodan Zhai, Dalin Guo, Wenzhe Shi, Jonathan J Hunt

Twitter developed a model-based reinforcement learning system to optimize push notification decisions by maximizing the discounted sum of future opened notifications. This system significantly reduced the number of notifications sent while increasing their open rate by 7.96% and slightly improving Daily Active Users (DAU) by 0.20% in a large-scale A/B experiment.

Key insights

  • Optimizing for long-term user value in push notifications, rather than immediate engagement, allows for sending fewer but more relevant messages, thereby improving user experience without sacrificing overall platform engagement.
  • A principled, data-driven model-based RL system offers a robust and adaptive alternative to static heuristic filters for managing complex recommender logic in large-scale production environments.
  • Incorporating dynamic user states, such as a 'streak' of consecutive opens or ignores, is crucial for modeling evolving user engagement and making more informed sequential decisions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.