LLM-Derived Priors for Thompson Sampling in Cold-Start Comment Recommendation

2026-08-04

AI Summary

This work addresses the poor recommendation performance of multi-armed bandits during the cold-start phase, when user interaction data are sparse. To mitigate this issue, the authors propose a personalized warm-start mechanism that integrates semantic priors derived from large language models (LLMs). Specifically, semantic signals extracted from user reviews are used to construct Bayesian priors, while posterior distributions for Thompson sampling are maintained separately across gender–age demographic segments. This approach represents the first integration of LLM-generated semantic priors into cold-start recommendation. Online A/B/C experiments demonstrate that the method significantly improves recommendation performance under sparse feedback conditions. Notably, gender-informed priors exhibit the strongest alignment with user click behavior, and heterogeneous treatment effects across demographic groups are statistically significant.

Abstract

Multi-armed bandit algorithms, especially Thompson sampling, are widely used in online recommendation. Despite their ability to adapt from online feedback, these methods often suffer from cold-start limitations when newly introduced arms have little or no interaction history. In our setting, the candidate arms are user-generated textual comments, whose semantic content can reveal a title's appeal before sufficient interaction feedback is available. We therefore use large language models (LLMs) to extract semantic signals from comment text and convert them into informative Bayesian priors that warm-start Thompson sampling under sparse early-stage feedback. To account for aggregate segment-level differences in response patterns, we maintain and update posteriors separately for each gender-age segment. In a real-world online A/B/C test, we compare a uniform prior with two LLM-based designs: a Gender Prior for demographic-affinity cues and a Content Prior for title-specific identity cues. The results show that LLM-based priors are most beneficial in sparse-feedback regimes -- with the largest gains emerging once a small amount of interaction evidence has accumulated -- and that prior design leads to distinct funnel-level effects. We further analyze prior-reward alignment and demographic heterogeneity, finding that click-oriented alignment is strongest for the Gender Prior and that treatment effects vary substantially across demographic segments. These findings suggest that LLM-derived priors can serve as a practical warm-start mechanism for text-rich bandit recommendation, while also revealing deployment trade-offs.

Problem

• cold-start
• Thompson sampling
• comment recommendation
• multi-armed bandit
• online recommendation

Innovation

• LLM-derived priors
• Thompson sampling
• cold-start recommendation
• multi-armed bandits
• demographic heterogeneity