Reproduce “Shape Your Feed: An LLM-based Agentic System for Conversational 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
Ziyun Xu, Bosen Ding, Yue Zhang, Ji Qi, Qingyuan Song, Jizhou Huang, Liwei Wang, Jefferey Santelli and 5 more
Meta Platforms introduces Shape Your Feed (SYF), an LLM-based agentic system that integrates user-explicit preferences into content recommendation feeds, moving beyond passive ranking. The system achieves a 1.70% reduction in post dismiss rates and a 2.74% drop in post dislike rates, alongside a 0.16% uplift in new interest consumption, by enabling real-time, multi-modal co-curation.
Key insights
- Integrating LLMs as structured controllers allows for effective interpretation of user intent and maintenance of dialogue context, enabling precise preference matching.
- A dual-feedback policy alignment, combining sparse live behavioral feedback with an LLM-as-a-Judge ensemble, is effective for scalable fine-tuning of LLM components under production constraints.
- The algorithmic re-ranking and candidate augmentation within the Serving Flow are critical for effectively executing user signals and improving feed quality, beyond just improved signal collection via new UI.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.