PsPLUG: A Lightweight Plug-in for Balancing Personalization and Style in Customized LLMs

Jul 16, 2026

Upload by Anda

Personalized large language models are often required to obey explicit style instructions, but explicit control can interfere with the user-specific characteristics that personalization techniques aim to retain. This failure mode is identified as personalization collapse, where style constraints conflict with implicit user preferences. PsPLUG is proposed as a lightweight plug-in that learns a user-specific residual after removing the requested style. Experiments show improved preference preservation and precise inference-time control over the balance between personalization and style adherence.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

Introduction

Personalization Collapse under Explicit Style Instructions

Residual View of Personalization

Style-Aware Preference Objective

PsPLUG Method and Inference-Time Control

Do Implicit Personalization and Explicit Styles Conflict? PsPLUG: A Lightweight Plug-in for Balancing Personalization and Style in Customized LLMs Yutong Song♠ * , Jiang Wu♣ , Shaofan Yuan♢ , Chengze Shen♢ , Jian Wang♢ , Yu Wang♢ , Nikhil Dutt♠ , Amir Rahmani♠♠University of California, Irvine ♣Independent Researcher ♢TikTok arXiv :2601 .06362v2 [ cs .AI] 13 Jul 2026 Abstract Personalized large language models are often expected to follow explicit style instructions, yet we find that such instructions can undermine the user-specific characteristics that personalization methods aim to preserve. We call this failure mode personalization collapse: explicit style control can conflict with implicit user preferences. To address this challenge, we propose PsPLUG, a lightweight plug-in that learns a user-specific residual after accounting for the requested style. PsPLUG also allows us to tune personalization strength at inference time. Our experiments show that explicit style instructions can diminish personalization in existing methods, whereas PsPLUG better preserves user preferences while providing precise control over the balance between personalization and style adherence. 1 Introduction Large language models (LLMs) are increasingly deployed in interactive settings where users expect systems not only to produce factually correct content, but also to align with their individual linguistic habits, preferences, and communicative styles (Tan and Jiang, 2023 ; Zhang et al., 2024) . This has sparked rapid progress in personalized generation, including (i) retrieval-based methods that fetch user histories into the context window (Salemi et al., 2024b ; Kumar et al., 2024 ; Mysore et al., 2024 ; Tang et al., 2024), (ii) peruser fine-tuning approaches such as user-specific LoRA or adapters (Hu et al., 2021 ; Houlsby et al., 2019 ; Tan et al., 2024a,b), and (iii) lightweight plug-in mechanisms that inject user embeddings or soft prompts (Liu et al., 2024 ; Li and Liang, 2021 ; Liu et al., 2022) . Together, these techniques have demonstrated that LLMs can adapt to a user given The code is available at [https://anonymous.4open](https://anonymous.4open). science/status/PsPLUG-038C Figure 1: The style-constrained personalization challenge. While standard models capture implicit user personalization, existing methods suffer from severe personalization collapse under explicit style instructions. enough data or context. (Chen et al., 2024, 2025b ; Zhang et al., 2025a) Despite this progress, a critical vulnerability in current personalization pipelines remains largely overlooked. Existing methods typically inject user-related signals without theoretically clarifying what constitutes the core personalized signal, or how it relates to the neutral behavior of the base model under the same inputs (Shenfeld et al., 2025 ; Kim et al., 2025) . Concurrently, modern NLP applications increasingly operate under explicit system instructions, such as strict stylistic or tonal guidance (e.g.,“respond formally”,“use a concise tone”) (Zhang et al., 2023 ; Liang et al., 2024 ; Shanahan et al., 2023) . We empirically observe that when such explicit constraints are introduced, existing personalization methods suffer from severe persona degradation. Strong system instructions tend to dominate the generation space, effectively overriding and collapsing diverse dimensions of user-specific traits. This reveals a fundamental challenge: style-constrained personalization, where a model must reliably balance explicit task directives with implicit user priors. To address this, we introduce a novel theoretical perspective: modeling personalization as a distribu- tional residual. Rather than learning absolute output likelihoods, we view the persona as the distinct deviation between two conditional distributions under the identical input: the user’s true linguistic distribution and the neutral distribution of a base LLM. A user-authored response natural

FAQs

What is “personalization collapse” in customized LLMs?

Personalization collapse is the failure mode where explicit style instructions override implicit user-specific preferences, causing user traits to degrade during generation.

How does PsPLUG preserve user personalization when style instructions are present?

PsPLUG learns a user-specific residual after accounting for the requested style, isolating persona signals from instruction-dominant effects.

How can the trade-off between style adherence and personalization strength be controlled?

PsPLUG introduces a unified inference-time scaling mechanism controlled by a coefficient (α), enabling dynamic adjustment without per-user fine-tuning.

Jul 16, 2026

Upload by Anda