Styles + Persona-plug = Customized LLMs

TL;DR

This paper introduces PsPLUG, a lightweight soft-prompt plugin that enhances personalized text generation by balancing implicit personalization with explicit style instructions, improving style alignment and fidelity.

Contribution

It formulates personalization as a distributional residual and proposes PsPLUG, a novel style-conditioned soft prompt method for controllable LLM personalization.

Findings

Improves persona alignment in style-conditioned text generation

Maintains stylistic fidelity with minimal computation

Outperforms retrieval-based and soft-prompt baselines

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.