InstructAgent: Building User Controllable Recommender via LLM Agent

2025-02-20

AI Summary

Traditional recommender systems adopt a platform-centric paradigm, where algorithms prioritize commercial objectives over individual user preferences—leading to diminished user agency, algorithmic manipulation, filter bubbles, and poor cold-start performance. This paper proposes a novel “user–agent–platform” paradigm, introducing an instruction-tunable recommendation agent powered by large language models (LLMs) as an intermediary that explicitly empowers users with control and decision authority over the recommendation process. Key contributions include: (1) the first instruction-driven recommendation agent architecture enabling explicit user command execution; (2) the construction of the first recommendation dataset annotated with explicit user-intent instructions; and (3) a structural shift of recommendation authority from platform to user. Extensive experiments on four newly curated datasets demonstrate significant improvements in perceived user control, personalized satisfaction, robustness against manipulation, mitigation of filter bubbles, and cold-start recommendation effectiveness.

Technology Category

• Data Mining & Knowledge Management: Recommender Systems
• Multiagent Systems: Mechanism Design
• Humans and AI: Intelligent User Interfaces

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Traditional recommender systems usually take the user-platform paradigm, where users are directly exposed under the control of the platform's recommendation algorithms. However, the defect of recommendation algorithms may put users in very vulnerable positions under this paradigm. First, many sophisticated models are often designed with commercial objectives in mind, focusing on the platform's benefits, which may hinder their ability to protect and capture users' true interests. Second, these models are typically optimized using data from all users, which may overlook individual user's preferences. Due to these shortcomings, users may experience several disadvantages under the traditional user-platform direct exposure paradigm, such as lack of control over the recommender system, potential manipulation by the platform, echo chamber effects, or lack of personalization for less active users due to the dominance of active users during collaborative learning. Therefore, there is an urgent need to develop a new paradigm to protect user interests and alleviate these issues. Recently, some researchers have introduced LLM agents to simulate user behaviors, these approaches primarily aim to optimize platform-side performance, leaving core issues in recommender systems unresolved. To address these limitations, we propose a new user-agent-platform paradigm, where agent serves as the protective shield between user and recommender system that enables indirect exposure. To this end, we first construct four recommendation datasets, denoted as $dataset$, along with user instructions for each record.

Problem

• Enhances user control over recommender systems
• Reduces platform manipulation and echo chambers
• Personalizes recommendations for less active users

Innovation

• LLM agent shields users
• Indirect exposure enhances control
• Custom datasets with instructions