UniSearch: Rethinking Search System with a Unified Generative Architecture Conference: Proceedings of the xxx; Beijing; China the Preprint Version of UniSearch, Beijing, China DOI: 10.1145/xxxxx ISBN: xxxxx/2025/09 CCS: Information systems Novelty in information retrieval CCS: Information systems Co
Jiahui Chen † , Xiaoze Jiang ∗† , Zhibo Wang † , Quanzhi Zhu † , Junyao Zhao † , Feng Hu † , Kang Pan, Ao Xie, Maohua Pei, Zhiheng Qin, Hongjing Zhang, Zhixin Zhai, Xiaobo Guo, Runbin Zhou, Kefeng Wang, Mingyang Geng, Cheng Chen, Jingshan Lv, Yupeng Huang, Xiao Liang, Han Li · 2025
Read UniSearch: Rethinking Search System with a Unified Generative Architecture Conference: Proceedings of the xxx; Beijing; China the Preprint Version of UniSearch, Beijing, China DOI: 10.1145/xxxxx ISBN: xxxxx/2025/09 CCS: Information systems Novelty in information retrieval CCS: Information systems Co by Jiahui Chen † , Xiaoze Jiang ∗† , Zhibo Wang † , Quanzhi Zhu † , Junyao Zhao † , Feng Hu † , Kang Pan, Ao Xie, Maohua Pei, Zhiheng Qin, Hongjing Zhang, Zhixin Zhai, Xiaobo Guo, Runbin Zhou, Kefeng Wang, Mingyang Geng, Cheng Chen, Jingshan Lv, Yupeng Huang, Xiao Liang, Han Li (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-06887.txt.
Description
Search software rebuilt as one generative system is presented by Jiahui Chen and twenty co-authors in a 2025 information retrieval paper. Traditional search engines chain recall, pre-ranking, and ranking modules, and that complexity blocks overall gains. Generative recommendation suggested an alternative, but earlier attempts trained the item tokenizer and the generator separately, causing mismatched goals.
UniSearch, built for Kuaishou Search, joins a search generator with a video encoder in one end-to-end architecture. The generator produces semantic identifiers of relevant items for a query, while the encoder learns item embeddings and their token forms. Joint training lets the parts improve each other. A preference optimization step adds a reward model and real user feedback. Industrial-scale experiments and online A/B tests in short-video and live search showed strong results, including a standout improvement in live search.
Catalog record
| CREATOR | Jiahui Chen † , Xiaoze Jiang ∗† , Zhibo Wang † , Quanzhi Zhu † , Junyao Zhao † , Feng Hu † , Kang Pan, Ao Xie, Maohua Pei, Zhiheng Qin, Hongjing Zhang, Zhixin Zhai, Xiaobo Guo, Runbin Zhou, Kefeng Wang, Mingyang Geng, Cheng Chen, Jingshan Lv, Yupeng Huang, Xiao Liang, Han Li |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-06887