CURE:Circuit-Aware Unlearning for LLM-based Recommendation

2026-04-04

AI Summary

This work addresses the challenge of data unlearning in large language model (LLM)-based recommender systems, where conventional uniform parameter updates often induce gradient conflicts between forgetting unwanted data and preserving useful knowledge, leading to significant performance degradation. To resolve this, the authors propose CURE, a novel framework that introduces causal circuit analysis into the unlearning process for LLM recommenders. CURE identifies computation subgraphs causally linked to recommendation tasks and partitions the model into three distinct modules: forget-specific, retain-specific, and task-shared. It then applies objective-aware, differential parameter updates tailored to each module’s role. This approach enables module-level functional disentanglement and a transparent, controllable update mechanism. Extensive experiments on real-world datasets demonstrate that CURE substantially outperforms existing baselines, achieving efficient data unlearning while better preserving recommendation accuracy.

Abstract

Recent advances in large language models (LLMs) have opened new opportunities for recommender systems by enabling rich semantic understanding and reasoning about user interests and item attributes. However, as privacy regulations tighten, incorporating user data into LLM-based recommendation (LLMRec) introduces significant privacy risks, making unlearning algorithms increasingly crucial for practical deployment. Despite growing interest in LLMRec unlearning, most existing approaches formulate unlearning as a weighted combination of forgetting and retaining objectives while updating model parameters in a uniform manner. Such formulations inevitably induce gradient conflicts between the two objectives, leading to unstable optimization and resulting in either ineffective unlearning or severe degradation of model utility. Moreover, the unlearning procedure remains largely black-box, undermining its transparency and trustworthiness. To tackle these challenges, we propose CURE, a circuit-aware unlearning framework that disentangles model components into functionally distinct subsets and selectively updates them. Here, a circuit refers to a computational subgraph that is causally responsible for task-specific behaviors. Specifically, we extract the core circuits underlying item recommendation and analyze how individual modules within these circuits contribute to the forget and retain objectives. Based on this analysis, these modules are categorized into forget-specific, retain-specific, and task-shared groups, each subject to function-specific update rules to mitigate gradient conflicts during unlearning. Experiments on real-world datasets show that our approach achieves more effective unlearning than existing baselines.

Problem

• LLM-based recommendation
• machine unlearning
• privacy
• gradient conflict
• model utility

Innovation

• circuit-aware unlearning
• LLM-based recommendation
• gradient conflict mitigation
• model disentanglement
• machine unlearning