Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Efficient and Effective Unlearning of Large Language Models for Recommendation

Hangyu Wang, Jianghao Lin, Bo Chen, Yang Yang, Ruiming Tang, Weinan Zhang, Yong Yu

TL;DR

This paper tackles the challenge of recommendation unlearning in large language model-based recommenders (LLMRec) by introducing E2URec, a method that achieves both efficiency and effectiveness. Efficiency is gained by inserting lightweight LoRA modules and only updating a small subset of parameters, while effectiveness is ensured via a two-teacher framework that guides forgetting on forgotten data and preserves performance on retained data. The forgetting teacher simulates retraining signals through logit differences, and the remembering teacher (the original model) anchors the retained data performance, with a combined loss $L = \beta L_{FGT} + (1-\beta) L_{REM}$. Extensive experiments on MovieLens-1M and GoodReads demonstrate that E2URec outperforms baselines in forgetting quality and recommendation performance, and scales across different LLM backbones, highlighting its practical impact for privacy-preserving, private-data-aware LLMRec systems.

Abstract

The significant advancements in large language models (LLMs) give rise to a promising research direction, i.e., leveraging LLMs as recommenders (LLMRec). The efficacy of LLMRec arises from the open-world knowledge and reasoning capabilities inherent in LLMs. LLMRec acquires the recommendation capabilities through instruction tuning based on user interaction data. However, in order to protect user privacy and optimize utility, it is also crucial for LLMRec to intentionally forget specific user data, which is generally referred to as recommendation unlearning. In the era of LLMs, recommendation unlearning poses new challenges for LLMRec in terms of \textit{inefficiency} and \textit{ineffectiveness}. Existing unlearning methods require updating billions of parameters in LLMRec, which is costly and time-consuming. Besides, they always impact the model utility during the unlearning process. To this end, we propose \textbf{E2URec}, the first \underline{E}fficient and \underline{E}ffective \underline{U}nlearning method for LLM\underline{Rec}. Our proposed E2URec enhances the unlearning efficiency by updating only a few additional LoRA parameters, and improves the unlearning effectiveness by employing a teacher-student framework, where we maintain multiple teacher networks to guide the unlearning process. Extensive experiments show that E2URec outperforms state-of-the-art baselines on two real-world datasets. Specifically, E2URec can efficiently forget specific data without affecting recommendation performance. The source code is at \url{https://github.com/justarter/E2URec}.

Towards Efficient and Effective Unlearning of Large Language Models for Recommendation

TL;DR

This paper tackles the challenge of recommendation unlearning in large language model-based recommenders (LLMRec) by introducing E2URec, a method that achieves both efficiency and effectiveness. Efficiency is gained by inserting lightweight LoRA modules and only updating a small subset of parameters, while effectiveness is ensured via a two-teacher framework that guides forgetting on forgotten data and preserves performance on retained data. The forgetting teacher simulates retraining signals through logit differences, and the remembering teacher (the original model) anchors the retained data performance, with a combined loss . Extensive experiments on MovieLens-1M and GoodReads demonstrate that E2URec outperforms baselines in forgetting quality and recommendation performance, and scales across different LLM backbones, highlighting its practical impact for privacy-preserving, private-data-aware LLMRec systems.

Abstract

The significant advancements in large language models (LLMs) give rise to a promising research direction, i.e., leveraging LLMs as recommenders (LLMRec). The efficacy of LLMRec arises from the open-world knowledge and reasoning capabilities inherent in LLMs. LLMRec acquires the recommendation capabilities through instruction tuning based on user interaction data. However, in order to protect user privacy and optimize utility, it is also crucial for LLMRec to intentionally forget specific user data, which is generally referred to as recommendation unlearning. In the era of LLMs, recommendation unlearning poses new challenges for LLMRec in terms of \textit{inefficiency} and \textit{ineffectiveness}. Existing unlearning methods require updating billions of parameters in LLMRec, which is costly and time-consuming. Besides, they always impact the model utility during the unlearning process. To this end, we propose \textbf{E2URec}, the first \underline{E}fficient and \underline{E}ffective \underline{U}nlearning method for LLM\underline{Rec}. Our proposed E2URec enhances the unlearning efficiency by updating only a few additional LoRA parameters, and improves the unlearning effectiveness by employing a teacher-student framework, where we maintain multiple teacher networks to guide the unlearning process. Extensive experiments show that E2URec outperforms state-of-the-art baselines on two real-world datasets. Specifically, E2URec can efficiently forget specific data without affecting recommendation performance. The source code is at \url{https://github.com/justarter/E2URec}.
Paper Structure (16 sections, 7 equations, 2 figures, 5 tables)

This paper contains 16 sections, 7 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Machine Unlearning and Downstream Applications
  4. Large Language Models as Recommenders
  5. Preliminary
  6. LLMs as Recommenders
  7. Recommendation Unlearning for LLMRec
  8. Methodologies
  9. Parameter-Efficient Unlearning
  10. Unlearning with Forgetting/Remembering Teachers
  11. Experiments
  12. Experiment Setups
  13. Results Analysis
  14. Impact of LLM
  15. Ablation Study
  16. ...and 1 more sections

Figures (2)

  • Figure 1: The illustration of recommendation unlearning. We aim to forget the recommendation knowledge about the chosen forgotten data due to privacy or utility purposes, and meanwhile keep the recommendation performance of the unlearned model for the rest data.
  • Figure 2: The overall framework of our proposed E2URec, which consists of two major stages: (1) Teacher Construction, and (2) Unlearning with Teachers.