Learning User Interests via Reasoning and Distillation for Cross-Domain News Recommendation

TL;DR

This paper introduces a reinforcement learning approach using large language models to infer deeper user interests from cross-domain signals, improving news recommendation relevance and scalability.

Contribution

It proposes a novel RL framework with query-list generation as policy optimization, and employs distillation for scalable deployment in news recommendation systems.

Findings

Consistent improvements in interest modeling quality.

Enhanced downstream recommendation performance.

Effective distillation from large models to compact ones.

Abstract

News recommendation plays a critical role in online news platforms by helping users discover relevant content. Cross-domain news recommendation further requires inferring user's underlying information needs from heterogeneous signals that often extend beyond direct news consumption. A key challenge lies in moving beyond surface-level behaviors to capture deeper, reusable user interests while maintaining scalability in large-scale production systems. In this paper, we present a reinforcement learning framework that trains large language models to generate high-quality lists of interest-driven news search queries from cross-domain user signals. We formulate query-list generation as a policy optimization problem and employ GRPO with multiple reward signals. We systematically study two compute dimensions: inference-time sampling and model capacity, and empirically observe consistent improvements with increased compute that exhibit scaling-like behavior. Finally, we perform on-policy distillation to transfer the learned policy from a large, compute-intensive teacher to a compact student model suitable for scalable deployment. Extensive offline experiments, ablation studies and large-scale online A/B tests in a production news recommendation system demonstrate consistent gains in both interest modeling quality and downstream recommendation performance.