Federated agents share knowledge for better personalized recommendations
Federated Recommender Agents
Information Retrieval
Summary
Recommender systems suggest items users might like, but sharing personal data raises privacy concerns. The paper explores a way where small agents on users' devices share advice using natural language, without exposing private information. The authors created FedRecAgent, which combines users' experiences by organizing them into condition, transition, and reasoning units to improve recommendations. Their approach helps agents learn from each other while keeping user data private, resulting in better personalized suggestions.
What this means in practice
- •For mobile app developers: Build personalized recommendation features that respect user privacy by enabling local reasoning agents to share knowledge without sharing personal data.
- •For consumer electronics teams: Design smart devices that improve user guidance and suggestions by integrating federated recommender agents communicating in natural language.
Authors
Jaehyung Lim, Woojoo Kim, Junyoung Kim, Dongha Kim, Wonbin Kweon, Hwanjo Yu
Abstract
LLM-powered recommender agents use reasoning and memory for personalization, but centralized access to user information raises privacy concerns, while isolated agents cannot learn from others' experiences. We formulate Federated Recommender Agents, a setting where local agents collaborate through natural-language knowledge while keeping interaction histories and personal memories local. We propose FedRecAgent, which enables this collaboration without updating LLM parameters. Its central challenge is determining which reasoning to integrate, since similar behavioral contexts can lead to different choices. FedRecAgent represents local experiences as Condition-Transition-Reasoning (C-T-R) units and hierarchically integrates reasoning by condition and transition. Within this structure, it consolidates related explanations and discriminates reasoning across alternative transitions. Each client maintains semantic and decision memories to capture preferences and conditional choice tendencies, guiding the retrieval and selection of shared reasoning for recommendation. Experiments across three domains under dense and sparse settings demonstrate improved recommendation performance over federated and LLM-based baselines, while ablations support the contributions of personalized memories and structured reasoning integration.