Reproduce “A Self-Triggered Agentic Push Recommendation System” 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

Zhao-Yu Zhang, Qingying Chen, Chunyuan Zheng, Jing Zhou, Jian Sun, Siqi Chen, Leiying Chen, Chuan Zhou and 4 more

Researchers from ByteDance and Peking University developed STEPS, a self-triggered agentic system for push recommendations that dynamically decides when to activate and whether to send a notification. Deployed on Douyin, the system increased user active days by 0.2843% and reduced push permission disablement by 1.9089%, while cutting computational overhead by 79.42%.

Key insights

  • A closed-loop, self-triggered agentic system can overcome limitations of passive, pre-defined schedules by dynamically optimizing both the timing and content of push notifications.
  • Combining a Gated-RTG mechanism with ordinal regression effectively addresses the challenges of noisy reward signals and non-Gaussian time distributions for robust planning.
  • A lightweight, item-feature-agnostic filtering agent is crucial for large-scale industrial deployment, significantly reducing computational overhead without sacrificing high-value user engagement opportunities.

Summary by alphaXiv.

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