Reproduce “TubiFM: Unified Item, Carousel, and Search Ranking for Streaming Discovery” 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
Alexandre Salle, Chenglei Niu, Suchismit Mahapatra, Xiaoxiao Chen, Suvash Sedhain, Yaqi Wang, Shervin Shahryari, Saurabh Agrawal and 2 more
TubiFM presents a unified generative ranking model that processes a serialized "user story" to perform item, carousel, and search ranking for streaming content. The model achieved improved performance in search and carousel ranking during online A/B tests and reduced P99 ranking latency across all tasks.
Key insights
- Unifying diverse ranking tasks through a comprehensive, serialized user story allows the model to effectively leverage complementary behavioral signals across different interaction types.
- A finetuned general-purpose large language model can be successfully adapted as a foundation model for complex, domain-specific recommendation and search tasks.
- Explicitly encoding session boundaries and temporal context within the user story significantly contributes to improved ranking performance for item, carousel, and search tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.