Reproduce “Baseline and Incremental: User, Item, and Request Debiasing in Instagram Feed Ranking” 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

Hsin-Chiang Chen, Hao Wang, Gedi Zhou, Mingze Gao, Abhishek Kumar, Yiben Yang, Haoyu Wang, Hongjing Xia and 18 more

Instagram Feed’s ranker scores candidate posts using engagement predictions that combine viewer, item or request baselines with candidate-specific response. The authors retain the raw prediction and add a separately weighted baseline-relative term using existing scoring-time signals. In a live four-arm test of six terms, both had higher point estimates than each single-term arm in 11/13 pairs, but arms were not retuned.

Summary by alphaXiv.

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