Search papers, labs, and topics across Lattice.
The paper introduces UniNote, a unified embedding model for industrial item-to-item (I2I) retrieval that addresses the limitations of existing multimodal embedding methods in balancing global representation with fine-grained local retrieval. UniNote employs tailored retrieval strategies for multimodal content and a two-stage training paradigm involving contrastive SFT and reinforcement learning to align the model with content relevance. Experiments demonstrate state-of-the-art performance across diverse I2I tasks, with significant improvements in retrieval quality and cost efficiency when deployed at Xiaohongshu using Matryoshka Representation Learning (MRL).
UniNote's two-stage training, combining contrastive SFT and RL, leapfrogs existing multimodal embeddings, delivering SOTA item-to-item retrieval performance with improved cost efficiency in real-world deployments.
Item-to-Item (I2I) retrieval is a fundamental part of modern content platforms, supporting critical industrial workflows from recommendation engines to content auditing. While multimodal embedding methods have advanced general retrieval, they often falter in I2I scenarios due to the challenges of balancing global content representation with fine-grained local retrieval, the systemic inefficiency of decoupled embedding-and-ranking pipelines, and the inherent trade-offs between model precision and serving latency. To solve these issues, we propose \textbf{UniNote}, a unified embedding model designed for industrial I2I retrieval. Tailored retrieval strategies are introduced to support representation learning over complex, multimodal content at varying granularities. To operationalize these strategies, UniNote employs a two-stage training paradigm: the first stage leverages contrastive SFT to establish robust base embeddings, while the second stage refines ranking quality through a reinforcement learning (RL) process that aligns the model with content relevance. Our results show that UniNote achieves SOTA performance across diverse I2I tasks. Deployed at Xiaohongshu and integrated with Matryoshka Representation Learning (MRL), UniNote achieved significant improvements in retrieval quality and cost efficiency in large-scale applications.
