Styles + Persona-plug = Customized LLMs

2026-01-10

arXiv.org

AI Summary

Existing approaches to personalized text generation struggle to simultaneously maintain character consistency and style fidelity under explicit stylistic instructions. This work proposes a novel formulation that models personalization as a residual over the language model’s output distribution. To this end, we introduce PsPLUG, a lightweight soft prompt plug-in that integrates style-conditioned preference-based contrastive learning. This approach enables efficient personalization without compromising the original style control mechanism. Requiring minimal computational overhead, PsPLUG significantly outperforms both retrieval-based and soft prompt baselines on the LaMP benchmark, while simultaneously improving character alignment and stylistic consistency.

Technology Category

• Natural Language Processing: Generation
• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

We discover a previously overlooked challenge in personalized text generation: personalization methods are increasingly applied under explicit style instructions, yet their behavior under such constraints remains poorly understood. To balance implicit personalization and explicit style, we formulate personalization as a distributional residual and propose PsPLUG, a lightweight soft-prompt plug-in trained with style-conditioned preference contrasts. Across LaMP benchmark, our framework improves persona alignment, maintains stylistic fidelity, and outperforms retrieval-based and soft-prompt baselines with minimal computation. These results show that residual modeling provides a simple and principled foundation for controllable, style-aware LLM personalization.

Problem

• personalized text generation
• style instruction
• persona alignment
• stylistic fidelity
• large language models

Innovation

• distributional residual
• style-conditioned personalization
• soft-prompt plug-in
• preference contrast
• controllable LLM