New AI Trick Makes Netflix Picks Smarter Without the Costly AI Bill

Researchers introduced P3Rec, a framework for LLM-based recommendation. Large language models show strong semantic understanding and preference reasoning, but existing LLM-as-Enhancer methods distill that knowledge into lightweight recommenders from only one perspective — prior preference captures stable, consistent interests but offers limited guidance for the current decision, while posterior preference reveals target-relevant fine-grained interests but may lean too heavily on target clues. P3Rec instead extracts complementary prior and posterior preference reasoning from the user side, plus item-centric preference representations from item semantics and predecessor interactions, then progressively internalizes them into behavioral representations. It also uses interest entropy to characterize historical interest dispersion and adaptively calibrate the user representation before contrastive retrieval optimization — aiming at more complete preference reasoning while preserving efficient recommendation. Extensive experiments on multiple public datasets demonstrate its effectiveness.