[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86200-en":3,"doc-seo-86200-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},86200,1374391974564,"Clementine","https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002",8,"Research & Report","Prompt Generation Technical Report","Prompt Generation (PG) is presented as a high-level, tokenizer- and configuration-driven framework to decouple feature-processing logic from generative retrieval model architecture in industrial search and recommendation systems. Using two declarative JSON files as a single source of truth, PG maintains feature consistency across offline training and online serving. Features are organized by four types and assembled via three composable components, enabling faster training iteration, scenario-agnostic deployment, and unified online inference optimizations. Validated on Taobao Search, Taobao Recommendation, and five benchmarks, PG improves offline retrieval quality and delivers statistically significant A/B uplifts in production.","arXiv :2607 . 1 1326v 1 [ cs .IR] 13 Jul 2026  \nPrompt Generation Technical Report  \nTaobao Search Team  \nGenerative retrieval has become an increasingly adopted paradigm for industrial search, recommendation, and advertising systems, delivering significant online gains. Most existing work combines user behavior sequences with large language models (LLMs) to model user preferences. In practice, feature engineering remains critical to model effectiveness, yet its complexity slows offline iteration and makes online deployment heavy and hard to reuse, all under tight online latency budgets. The root cause is a tight coupling between feature-processing logic and model architecture, where every feature change touches the training and serving code and resists reuse across scenarios.  \nTo break this coupling, as shown in Figure 1, we present Prompt Generation (PG), a high-level tokenizerand configuration-driven framework that decouples feature-processing logic from model architecture through two declarative JSON files, which serve as the single source of truth for both offline training and online serving, ensuring feature consistency across the two stages. Organizing features under four types with three composable processing components to assemble and compress heterogeneous features, PG delivers acceleration at three levels:  \n(1) fast training iteration: feature experiments require only configuration changes, with built-intoken compression for ultra-long sequences;  \n(2) fast deployment: a new scenario only needs to conform to the PG schema and plug into a universal pipeline, with no scenario-specific engineering;  \n(3) fast online inference: engine applies unified optimizations over the standardized configuration, reducing PG’s overhead to a negligible level.  \nWe validate PG on Taobao Search, Taobao Recommendation, and three open-source benchmarks, where configuration adaptation alone improves offline retrieval quality across all five scenarios. In production, PG has been deployed on Taobao Search with statistically significant online A/B uplifts of +0 .47% in transaction count and +0 . 51% in GMV, and has been applied across multiple Taobao search and recommendation teams as the iteration framework for generative retrieval.  \nDataset  \nGR without PG   \n\n| Nick | Age | Sex | UEmb | Clicks | Label |\n| --- | --- | --- | --- | --- | --- |\n| Bob | 23 | M | [0.8] | C2,C3 | L1 |\n| ... | ... | ... | ... | ... | ... |\n\nFigure 1 | Generative Retrieval (GR) without vs. with Prompt Generation (PG) . PG decouples feature processing from model architecture, turning a static pipeline into a dynamic, model-agnostic one.  \nContents  \n1 Introduction 3  \n2 Protocol Design 4  \n2.1 Design Overview ...................................... 4  \n2.2 Feature Type Taxonomy .................................. 5  \n2.3 Configuration Schema ................................... 6  \n2.4 Configuration Example Walkthrough ........................... 13  \n3 System Architecture and Implementation 15  \n3. 1 Architecture Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15  \n3.2 Offline Training Pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16  \n3.3 Online Inference Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17  \n4 Experiments 19  \n4.1 Taobao Search Feature Experiments ............................ 19  \n4.2 Taobao Recommendation Feature Experiments ...................... 20  \n4.3 Open-source Benchmark Feature Experiments ...................... 21  \n4.4 Alignment Experiments ................................... 24  \n4.5 Latency Analysis ...................................... 25  \n4.6 Online Deployment ..................................... 29  \n5 Discussion 30  \n5.1 Key Findings ........................................ 30  \n5.2 Design Insights ....................................... 31  \n5.3 Integration with Autoresearch ............................... 31  \n5.4 Future Work . . . . . . ","cbCaikLGrC5dtvAd","https://ap.wps.com/l/cbCaikLGrC5dtvAd","pdf",1347853,1,35,"English","en",105,"# Introduction\n## Protocol Design\n## System Architecture and Implementation\n## Experiments\n## Discussion\n## Conclusion","[{\"question\":\"What problem does Prompt Generation (PG) solve in generative retrieval systems?\",\"answer\":\"PG addresses the tight coupling between feature-processing logic and model architecture, where changing features requires modifying training and serving code and limits reuse under latency constraints.\"},{\"question\":\"How does PG maintain feature consistency between offline training and online serving?\",\"answer\":\"PG uses two declarative JSON files as the single source of truth for both offline training and online serving, ensuring the same feature definitions are applied across stages.\"},{\"question\":\"What performance benefits does PG deliver in practice?\",\"answer\":\"Configuration adaptation improves offline retrieval quality across all tested scenarios, and production deployment on Taobao Search shows statistically significant online A/B uplifts in transaction count and GMV.\"}]",1784209368,88,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"prompt-generation-technical-report","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/prompt-generation-technical-report/86200/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does Prompt Generation (PG) solve in generative retrieval systems?","Question",{"text":75,"@type":76},"PG addresses the tight coupling between feature-processing logic and model architecture, where changing features requires modifying training and serving code and limits reuse under latency constraints.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does PG maintain feature consistency between offline training and online serving?",{"text":80,"@type":76},"PG uses two declarative JSON files as the single source of truth for both offline training and online serving, ensuring the same feature definitions are applied across stages.",{"name":82,"@type":73,"acceptedAnswer":83},"What performance benefits does PG deliver in practice?",{"text":84,"@type":76},"Configuration adaptation improves offline retrieval quality across all tested scenarios, and production deployment on Taobao Search shows statistically significant online A/B uplifts in transaction count and GMV.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]