Interactive Recommendation Agent with Active User Commands

2025-09-25

AI Summary

Traditional recommender systems rely on passive implicit feedback (e.g., clicks, ratings), yielding coarse-grained, attribution-ambiguous signals that hinder precise user preference modeling and sustain a persistent gap between user intent and system interpretability. To address this, we propose Interactive Recommendation Framework (IRF), the first recommendation paradigm introducing natural-language-instruction-driven active interaction, enabling users to dynamically steer recommendation strategies in real time. Methodologically, IRF employs a dual-agent architecture—Parser Agent converts instructions into structured preferences, while Planner Agent orchestrates toolchains adaptively—augmented by simulation-enhanced knowledge distillation to improve both inference efficiency and transparency. Offline evaluations and long-term online A/B tests demonstrate statistically significant improvements in user satisfaction and core business metrics (e.g., CTR, dwell time, conversion rate), empirically validating the effectiveness and practical viability of active interaction in modern recommender systems.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Active Learning
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Assisted, interactive, and conversational search
• 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 rely on passive feedback mechanisms that limit users to simple choices such as like and dislike. However, these coarse-grained signals fail to capture users' nuanced behavior motivations and intentions. In turn, current systems cannot also distinguish which specific item attributes drive user satisfaction or dissatisfaction, resulting in inaccurate preference modeling. These fundamental limitations create a persistent gap between user intentions and system interpretations, ultimately undermining user satisfaction and harming system effectiveness. To address these limitations, we introduce the Interactive Recommendation Feed (IRF), a pioneering paradigm that enables natural language commands within mainstream recommendation feeds. Unlike traditional systems that confine users to passive implicit behavioral influence, IRF empowers active explicit control over recommendation policies through real-time linguistic commands. To support this paradigm, we develop RecBot, a dual-agent architecture where a Parser Agent transforms linguistic expressions into structured preferences and a Planner Agent dynamically orchestrates adaptive tool chains for on-the-fly policy adjustment. To enable practical deployment, we employ simulation-augmented knowledge distillation to achieve efficient performance while maintaining strong reasoning capabilities. Through extensive offline and long-term online experiments, RecBot shows significant improvements in both user satisfaction and business outcomes.

Problem

• Traditional systems fail to capture nuanced user motivations and intentions
• Current methods cannot identify specific attributes driving user satisfaction
• There exists a persistent gap between user intentions and system interpretations

Innovation

• Natural language commands for recommendation feeds
• Dual-agent architecture with parser and planner
• Simulation-augmented knowledge distillation for deployment