CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation
This work addresses the problem of scalable and efficient personalized text generation for users, representing an incremental advancement by combining clustering and lightweight decoding techniques.
The paper tackles the challenge of personalizing large language models for individual users by proposing CARD, a hierarchical framework that clusters users and uses cluster-specific adapters with reward-guided decoding, achieving competitive or superior generation quality on LaMP and LongLaMP benchmarks while improving efficiency and scalability.
Adapting large language models to individual users remains challenging due to the tension between fine-grained personalization and scalable deployment. We present CARD, a hierarchical framework that achieves effective personalization through progressive refinement. CARD first clusters users according to shared stylistic patterns and learns cluster-specific LoRA adapters, enabling robust generalization and strong low-resource performance. To capture individual differences within each cluster, we propose an implicit preference learning mechanism that contrasts user-authored text with cluster-level generations, allowing the model to infer user-specific style preferences without manual annotation. At inference time, CARD injects personalization exclusively at decoding via lightweight user preference vectors and low-rank logit corrections, while keeping the base model frozen. Experiments on the LaMP and LongLaMP benchmarks show that CARD achieves competitive or superior generation quality compared to state-of-the-art baselines, while significantly improving efficiency and scalability for practical personalized text generation.