[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84923-en":3,"doc-seo-84923-105":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},84923,687197207639,"Asher","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","DynaKRAG: A Unified Framework for Learnable Evidence Control in Multi-Hop Retrieval-Augmented Generation","Multi-hop retrieval-augmented generation (RAG) acquires evidence sequentially, where each newly retrieved document can change the information need by revealing missing facts, bridge entities, query defects, or enough support to answer. DynaKRAG formulates multi-hop evidence acquisition as state-conditioned control over atomic evidence operations. A validity layer builds the executable action set each step, and a learned controller selects the next operation. With Qwen2.5-7B-Instruct, it surpasses strong controlled baselines on HotpotQA, 2Wiki, and MuSiQue, and ablations show the value of sufficiency feedback and cost-aware, state-evolving decision making.","DynaKRAG: A Unified Framework for Learnable Evidence Control in Multi-Hop  \nRetrieval-Augmented Generation  \nYaqi Wu 1 ∗ , Xiaolei Guo 1 ∗ , Chenyu Zhou 1 , Jiaqi Huang 1 , Xianfa Zhang2 , Junxu Zhang2 , Zhuo Yu2 ,  \nZhubo Shi3 , Jianghao Lin 1†, Dongdong Ge 1  \n1 Shanghai Jiao Tong University  \n2 Shanghai Aircraft Manufacturing Co., Ltd.  \n3Tongji University  \n{wuyaqi7, lionelgxl, chenyuzhou, linjianghao, [ddge}@sjtu.edu.cn](ddge}@sjtu.edu.cn)  \n[hjq122418@gmail.com](hjq122418@gmail.com), {zhangxianfa, zhangjunxu, yuzhuo, [shizhubo}@comac.cc](shizhubo}@comac.cc)  \narXiv :2607 .06507v 1 [ cs .CL] 7 Jul 2026  \nAbstract  \nMulti-hop retrieval-augmented generation (RAG) acquires evidence sequentially, with each new document potentially revealing missing facts, bridge entities, query defects, or sufficient support for answering. Existing methods provide useful operations such as iterative retrieval, query reformulation, evidence critique, and sufficiency judging, but typically organize them within method-specific pipelines or predefined control topologies. This leaves underexplored how to learn a shared state-conditioned policy that chooses among currently valid evidence operations. We introduce DynaKRAG, which formulates multi-hop evidence acquisition as state-conditioned control over atomic evidence operations. At each step, a validity layer constructs the executable action set, and a learned controller selects the next operation. The resulting transition updates the evidence state and may enable new operations at subsequent steps. With Qwen2.5-7B-Instruct, DynaKRAG achieves F1 scores of 0.5998 on HotpotQA, 0.5340 on 2Wiki, and 0.3061 on MuSiQue, outperforming the strongest controlled baseline on all three benchmarks. Replacing the learned controller with a uniform-valid policy reduces F1 by 3.96– 5.78 points, while removing sufficiency feedback hurts all three datasets. Controlled retrieval-cap experiments further show that additional retrieval is not uniformly beneficial. Together, these results demonstrate the benefit of coordinating retrieval, diagnosis, and gap-directed acquisition under an evolving evidence state.  \nIntroduction  \nRetrieval-augmented generation (RAG) grounds language models in external sources of evidence (Lewis et al. 2020), but multi-hop questions break the usual retrieve-thengenerate abstraction. The first useful passage rarely completes the answer; instead, it changes the information need. It may expose a bridge entity, reveal a missing relation, show that the current query is misdirected, or provide enough support to stop. The next step is therefore not just a decision about retrieval depth. It is a control decision over an evolving evidence state: whether to continue along the retrieval frontier, reformulate the query, expand around a bridge entity, request a missing fact, check sufficiency, or stop and answer.  \n∗These authors contributed equally.†Corresponding author.  \nFigure 1: DynaKRAG overcomes fragmented, methodspecific RAG pipelines by learning unified control over valid atomic evidence operations, enabling effective and tokenefficient evidence acquisition.  \nAdaptive RAG methods recognize parts of this problem by allowing intermediate results to guide later retrieval and reasoning (Trivedi et al. 2023; Jeong et al. 2024). Recent systems further introduce useful behaviors such as query reformulation, evidence critique, sufficiency judging, and gap-directed  \nretrieval (Jiang et al. 2023; Su et al. 2024; Asai et al. 2024; Yan et al. 2024; Li et al. 2026a,b) . Yet these behaviors are still largely packaged inside method-specific pipelines. Each pipeline fixes its own state representation, action schedule, and control topology, which makes heterogeneous evidence operations difficult to express, compare, and learn within one framework (Figure 1, top). This is more than an engineering inconvenience: a system that can decide when to retrieve more may not know when rewriting is better; a sufficiency ","cbCairVdga0t0L6H","https://ap.wps.com/l/cbCairVdga0t0L6H","pdf",1128829,1,10,"English","en",105,"# Abstract\n# Introduction\n## Motivation: Multi-hop evidence acquisition as control\n## Efficiency and cost-aware decision making\n## DynaKRAG framework overview\n## Key mechanism: validity vs utility","[{\"question\":\"What problem does DynaKRAG address in multi-hop retrieval-augmented generation?\",\"answer\":\"It targets how to choose among currently valid evidence operations as the evidence state evolves, instead of relying on fixed, method-specific control pipelines.\"},{\"question\":\"How does DynaKRAG decide the next action during evidence acquisition?\",\"answer\":\"A hard validity layer first constructs the executable action set for the current state, and a learned value model ranks only those valid choices to select the next operation.\"},{\"question\":\"What do the reported experiments show about DynaKRAG’s effectiveness?\",\"answer\":\"Using Qwen2.5-7B-Instruct, DynaKRAG achieves higher F1 scores than the strongest controlled baseline on HotpotQA, 2Wiki, and MuSiQue, and ablations indicate that removing the learned controller or sufficiency feedback degrades performance.\"}]",1784199364,25,{"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},"dynakrag-a-unified-framework-for-learnable-evidence-control-in-multi-hop-retrieval-augmented-generation","",{"@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/dynakrag-a-unified-framework-for-learnable-evidence-control-in-multi-hop-retrieval-augmented-generation/84923/",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 DynaKRAG address in multi-hop retrieval-augmented generation?","Question",{"text":75,"@type":76},"It targets how to choose among currently valid evidence operations as the evidence state evolves, instead of relying on fixed, method-specific control pipelines.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does DynaKRAG decide the next action during evidence acquisition?",{"text":80,"@type":76},"A hard validity layer first constructs the executable action set for the current state, and a learned value model ranks only those valid choices to select the next operation.",{"name":82,"@type":73,"acceptedAnswer":83},"What do the reported experiments show about DynaKRAG’s effectiveness?",{"text":84,"@type":76},"Using Qwen2.5-7B-Instruct, DynaKRAG achieves higher F1 scores than the strongest controlled baseline on HotpotQA, 2Wiki, and MuSiQue, and ablations indicate that removing the learned controller or sufficiency feedback degrades performance.","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},[93,97,101,105,110,115,120,123,128,131,134],{"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":21,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":21,"slug":133},"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]