Preprint
Aug 2026
LLM-Derived Priors for Thompson Sampling in Cold-Start Comment Recommendation
It is suggested that LLM-derived priors can serve as a practical warm-start mechanism for text-rich bandit recommendation, while also revealing deployment trade-offs.
E. Lee, Oseong Choi, Byungsoo Kang et al.
· 0 citations