NAVER Paper: Using LLM-Derived Bayesian Priors to Solve Cold-Start in Recommendation

_reachsumit · x · 2026-08-05

NAVER proposes a novel method using Large Language Models (LLMs) to address the cold-start problem in recommendation systems.

The research uses LLMs to extract semantic signals from user comments, converting them into Bayesian priors to warm-start Thompson sampling in multi-armed bandit algorithms. In a live A/B/C test, this LLM-based prior design showed the highest CTR gains in sparse-feedback regimes, with distinct funnel-level effects across different demographic segments.

Original post →

More from Research

Research channel →