OneSearch - Kuaishou's end-to-end generative framework for e-commerce search.

What is OneSearch?

OneSearch is an end-to-end generative framework for e-commerce search launched by Kuaishou. It optimizes the cascaded architecture of traditional e-commerce search, improving search accuracy and user experience. Its three major innovations include: a Keyword Enhancement Hierarchical Quantization Encoding (KHQE) module, which strengthens query-product relevance constraints by extracting core product attributes and generating hierarchical codes (SIDs); a multi-perspective user behavior sequence injection strategy, which constructs behavior-driven user identifiers (UIDs), integrating explicit short-term behaviors and implicit long-term sequences to accurately model user preferences; and a Preference-Aware Reward System (PARS), which combines multi-stage supervised fine-tuning and adaptive reward reinforcement learning to capture fine-grained user preference signals. OneSearch significantly outperforms traditional systems in offline experiments, and in online experiments, it increases order volume by 3.22% and the number of buyers by 2.4%, demonstrating outstanding performance in long-tail queries and new product cold start scenarios.

OneSearch's main functions

• Precise matchingBy using the Keyword Enhanced Hierarchical Quantization (KHQE) module, the core attributes of products are accurately extracted and hierarchical codes are generated, significantly improving the distinguishability and accuracy of generative retrieval. It can understand colloquial, vague, and even incomplete expressions and transform them into efficient shopping instructions.
• Efficient sortingBy employing a multi-perspective user behavior sequence injection strategy, we construct behavior-driven user identifiers (UIDs), integrate explicit short-term behaviors with implicit long-term sequences, comprehensively and accurately model user preferences, and achieve more intelligent result ranking.
• Cost optimizationAfter going live, online inference costs decreased by 75.4%, machine computing efficiency increased by 8 times, and operating costs were significantly reduced.
• Improve user experienceIn human evaluations, OneSearch significantly outperformed traditional systems in overall page satisfaction, product quality, and query-item relevance, demonstrating a more comprehensive understanding of user intent and significantly improving the accuracy of personalized searches and user experience.
• Helping merchantsIt performs particularly well in cold-start scenarios, significantly outperforming conventional scenarios, indicating that generative retrieval models can more effectively address the ranking challenges of long-tail users and newly listed products.

OneSearch's technical principles

• Keyword Enhancement Hierarchical Quantization (KHQE)By extracting the core attributes of products, such as brand, category, color, and material, a hierarchical "smart ID" (SID) is generated for each product, thereby significantly improving the distinguishability and accuracy of generative retrieval.
• Multi-perspective user behavior sequence injection: Construct behavior-driven user identifiers (UIDs) that integrate explicit short-term behaviors with implicit long-term sequences to comprehensively and accurately model user preferences in order to achieve more intelligent result ranking.
• Preference-Aware Reward System (PARS)By combining multi-stage supervised fine-tuning (SFT) with adaptive reward reinforcement learning, fine-grained user preference signals are captured, enhancing the model's personalized ranking capabilities.

OneSearch's project address

• arXiv technical paper: https://arxiv.org/pdf/2509.03236

Application scenarios of OneSearch

• e-commerce searchOneSearch significantly enhances the user search experience on e-commerce platforms through accurate matching and efficient sorting, helping users find the products they need faster.
• Long-tail product recommendationsIn cold start scenarios, OneSearch can more effectively handle the sorting of long-tail users and newly listed products, increasing the exposure and sales opportunities of long-tail products.
• Personalized searchBy injecting user behavior sequences from multiple perspectives, OneSearch can accurately model user preferences and provide personalized search results to meet the needs of different users.
• Improve merchant operational efficiencyOneSearch helps merchants increase product exposure and conversion rates, and improve overall operational efficiency by optimizing search results.