#artificial intelligence
Jan 2026
CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation
This work presents CARD, a hierarchical framework that achieves effective personalization through progressive refinement that first clusters users according to shared stylistic patterns and learns group-specific LoRA adapters, enabling robust generalization and strong low-resource performance.
Yu-Tong Song, Jiang Wu, Weijia Zhang et al.
· arXiv.org · 3 citations