MRP-LLM: Multitask Reflective Large Language Models for Privacy-Preserving Next POI Recommendation

TL;DR

MRP-LLM addresses privacy-preserving next POI recommendation by integrating multitask reflective preference extraction, neighbor preference retrieval, and a privacy-aware data transmission pipeline. It distills fine-grained user preferences into a knowledge base, aggregates collaborative signals from neighboring users, and generates next-POI recommendations through multitask prompting, while keeping sensitive data locally and applying differential-privacy-based perturbations. Experiments on three real-world datasets show that MR-LLM improves over strong baselines and that the privacy-preserving variant (MRP-LLM) maintains competitive performance with a modest utility drop, achieving a favorable privacy-utility balance. The approach offers interpretable prompts and explanations, advancing privacy-preserving, LLM-based POI recommendation with collaborative signals.

Abstract

Large language models (LLMs) have shown promising potential for next Point-of-Interest (POI) recommendation. However, existing methods only perform direct zero-shot prompting, leading to ineffective extraction of user preferences, insufficient injection of collaborative signals, and a lack of user privacy protection. As such, we propose a novel Multitask Reflective Large Language Model for Privacy-preserving Next POI Recommendation (MRP-LLM), aiming to exploit LLMs for better next POI recommendation while preserving user privacy. Specifically, the Multitask Reflective Preference Extraction Module first utilizes LLMs to distill each user's fine-grained (i.e., categorical, temporal, and spatial) preferences into a knowledge base (KB). The Neighbor Preference Retrieval Module retrieves and summarizes the preferences of similar users from the KB to obtain collaborative signals. Subsequently, aggregating the user's preferences with those of similar users, the Multitask Next POI Recommendation Module generates the next POI recommendations via multitask prompting. Meanwhile, during data collection, a Privacy Transmission Module is specifically devised to preserve sensitive POI data. Extensive experiments on three real-world datasets demonstrate the efficacy of our proposed MRP-LLM in providing more accurate next POI recommendations with user privacy preserved.

Figures (9)

• Figure 1: Our proposed method shown in (b), which makes recommendations based on more diverse and fine-grained user preferences, compared to existing direct zero-shot methods shown in (a).
• Figure 2: The architecture of MRP-LLM. Taking $u_1$ as the recommendation query user.
• Figure 3: 〈$\varrho,h$〉-Privacy for POI check-ins
• Figure 4: Recommendation Process of MRP-LLM
• Figure 5: Results of ablation study.