SPRInG: Continual LLM Personalization via Selective Parametric Adaptation and Retrieval-Interpolated Generation

2026-01-15

AI Summary

This work proposes a semi-parametric continual personalization framework to address catastrophic forgetting, noise sensitivity, and personalization failure in large language models under continuously evolving user preferences. The approach identifies genuine preference shifts through novelty scoring and employs drift-driven selective adaptation coupled with a stringent relevance gating mechanism to effectively distinguish long-term preference changes from transient contextual cues. By integrating retrieved historical interactions with parametric knowledge, the framework leverages likelihood scoring, user-specific adapters, a replay buffer, and logit interpolation to enable robust generation. Evaluated on long-text personalization benchmarks, the method significantly outperforms existing approaches, demonstrating its effectiveness and robustness in dynamic interactive scenarios.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Natural Language Processing: (Large) Language Models
• Humans and AI: Learning Human Values and Preferences

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Personalizing Large Language Models typically relies on static retrieval or one-time adaptation, assuming user preferences remain invariant over time. However, real-world interactions are dynamic, where user interests continuously evolve, posing a challenge for models to adapt to preference drift without catastrophic forgetting. Standard continual learning approaches often struggle in this context, as they indiscriminately update on noisy interaction streams, failing to distinguish genuine preference shifts from transient contexts. To address this, we introduce SPRInG, a novel semi-parametric framework designed for effective continual personalization. During training, SPRInG employs drift-driven selective adaptation, which utilizes a likelihood-based scoring function to identify high-novelty interactions. This allows the model to selectively update the user-specific adapter on drift signals while preserving hard-to-learn residuals in a replay buffer. During inference, we apply strict relevance gating and fuse parametric knowledge with retrieved history via logit interpolation. Experiments on the long-form personalized generation benchmark demonstrate that SPRInG outperforms existing baselines, validating its robustness for real-world continual personalization.

Problem

• continual personalization
• preference drift
• catastrophic forgetting
• large language models
• dynamic user preferences

Innovation

• continual personalization
• selective parametric adaptation
• preference drift
• retrieval-interpolated generation
• semi-parametric framework