Daoxin Zhang
Head of Multimodal Search & Internationalization Algorithms at Xiaohongshu
Daoxin Zhang is currently the head of the Multimodal Search and Rednote Algorithm teams at Xiaohongshu. He holds a master's degree from Zhejiang University. He has long focused on multimodal understanding and retrieval, leading the development of various multimodal systems at Xiaohongshu and Alibaba, including visual image search, product understanding and same-item retrieval, as well as video structuring and intelligent production. He has extensive industry implementation experience and has published multiple papers in conferences such as ICCV, MM, and SIGIR. His main research areas include multimodal understanding and evaluation, video generation, and cross-modal retrieval.
Topic
Application of Multimodal Large Models in Xiaohongshu Search
With the rapid advancement of large model technologies, search has become one of the business systems most directly impacted. LLMs significantly enhance the modeling capabilities of traditional retrieval systems and, in the form of AI search, efficiently handle a portion of user queries. This talk will first introduce the overall business and architecture of Xiaohongshu Search, highlighting the scenarios and challenges faced by UGC-centric search. Focusing on multimodal search, the presentation will cover four directions: image-to-image search, image search, video search, and multimodal AI search, illustrating business scenarios and application progress for multimodal large models. Using typical application examples, it will delve into algorithmic details of multimodal LLMs in content understanding and RAG systems, and discuss the challenges and practical experiences of deploying multimodal LLMs in large-scale business environments.
