Real-Time Hard Negative Sampling via LLM-based Clustering for Large-Scale Two-Tower Retrieval

2026-07-01

AI Summary

This work addresses the limitations of conventional negative sampling strategies in two-tower model training, which often yield easy negatives that hinder discriminative learning and exacerbate popularity bias and feedback loops. To overcome these issues, the authors propose the first large language model (LLM)-based framework for real-time hard negative sampling. During training, the LLM performs semantic clustering and dynamically generates challenging yet relevant negatives from within semantically similar clusters. This approach significantly enhances the representation learning capability of two-tower models, effectively mitigates popularity bias, and breaks detrimental feedback cycles—all while maintaining low computational overhead. Extensive experiments demonstrate that the proposed framework consistently outperforms state-of-the-art industrial negative sampling methods on both public benchmarks and a billion-scale production system, yielding substantial gains in retrieval performance.

Abstract

The two-tower model has been widely used for large-scale recommendation systems, particularly in the retrieval stage. Industry standards for training two-tower models typically involve in-batch and/or out-of-batch negative sampling. However, these methods often produce easy negatives that models can quickly learn, failing to sufficiently challenge the model. To address this issue, a novel self-supervised hard negative sampling technique is proposed that leverages a large language model (LLM) to generate hard negatives from the same cluster during model training. By utilizing the LLM to learn media representations, the proposed approach ensures that the generated negatives are more challenging and informative. This real-time sampling framework is designed for seamless integration into production models, capable of handling billions of training data points with minimal computational complexity. Experiments on public datasets, along with deployment to a large-scale online system, demonstrate that the proposed negative sampling technique outperforms widely used industry methods. Furthermore, analysis in industrial applications reveals that this sampling method can help break inherent feedback loops in recommendations and significantly reduce popularity bias.

Problem

• hard negative sampling
• two-tower retrieval
• large-scale recommendation
• popularity bias
• feedback loops

Innovation

• hard negative sampling
• LLM-based clustering
• two-tower retrieval
• real-time sampling
• popularity bias reduction