STARec: An Efficient Agent Framework for Recommender Systems via Autonomous Deliberate Reasoning

TL;DR

STARec introduces a deliberative agent framework for recommender systems that enhances reasoning and decision-making, leading to improved performance with minimal training data.

Contribution

It presents a novel slow-thinking agent framework with anchored reinforcement training, enabling autonomous reasoning in recommender systems.

Findings

Significant performance improvements on MovieLens 1M and Amazon CDs datasets.

Achieves these gains using only 0.4% of the full training data.

Demonstrates the effectiveness of autonomous deliberative reasoning in recommendation accuracy.

Abstract

While modern recommender systems are instrumental in navigating information abundance, they remain fundamentally limited by static user modeling and reactive decision-making paradigms. Current large language model (LLM)-based agents inherit these shortcomings through their overreliance on heuristic pattern matching, yielding recommendations prone to shallow correlation bias, limited causal inference, and brittleness in sparse-data scenarios. We introduce STARec, a slow-thinking augmented agent framework that endows recommender systems with autonomous deliberative reasoning capabilities. Each user is modeled as an agent with parallel cognitions: fast response for immediate interactions and slow reasoning that performs chain-of-thought rationales. To cultivate intrinsic slow thinking, we develop anchored reinforcement training - a two-stage paradigm combining structured knowledge distillation from advanced reasoning models with preference-aligned reward shaping. This hybrid approach scaffolds agents in acquiring foundational capabilities (preference summarization, rationale generation) while enabling dynamic policy adaptation through simulated feedback loops. Experiments on MovieLens 1M and Amazon CDs benchmarks demonstrate that STARec achieves substantial performance gains compared with state-of-the-art baselines, despite using only 0.4% of the full training data.