🤖 AI Summary
This work addresses the high training costs, substantial inference latency, and poor deployment scalability prevalent in large-scale news recommendation systems by proposing a training-free, zero-parameter personalized recommendation framework. The approach achieves personalization through efficient matching between user behavioral representations and news semantics, entirely circumventing neural network training or fine-tuning. Experimental results demonstrate that the method outperforms strong neural baselines in offline evaluations, while online A/B tests show click-through rates nearly on par with state-of-the-art models. Notably, it achieves over a 600× speedup in inference latency, offering compelling evidence of the often-overlooked discrepancy between offline metrics and online performance. To the best of our knowledge, this is the first practical news recommendation system that simultaneously delivers high performance and eliminates the need for model training.
📝 Abstract
We present ZoRRO (Zero-Weight Personalized Recommender System), a zero-weight, training-free framework for personalized news recommendation designed for scalable real-world deployment. ZoRRO outperforms strong neural baselines in offline ranking evaluations and achieves click-through rate performance in online A/B testing that is nearly on par with a state-of-the-art deep learning model, while operating more than 600 times faster. Our experiments reveal gaps between offline and online performance and demonstrate that models with similar click-through rate outcomes can produce markedly different recommendation distributions, thereby influencing the overall news flow. These findings position ZoRRO as a practical and efficient solution for large-scale news recommendation and highlight the importance of evaluating recommender systems using metrics beyond accuracy alone.