[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85082-en":3,"doc-seo-85082-105":30,"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85082,1099514067438,"River Wang","https://ap-avatar.wpscdn.com/avatar/100002539ee87300030?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780474512215547542",8,"Research & Report","Game Theory Driven Multi-Agent Framework Mitigates Language Model Hallucination","Lightweight Large Language Models in rule-based scientific domains often imitate linguistic patterns instead of performing axiomatic reasoning, leading to frequent hallucinations. G-Frame, an adaptive multi-agent framework combining Bayesian and team game principles, creates an automated closed-loop for high-quality data synthesis and model training. Structured reasoning internalizes domain constraints and produces a specialized dataset with 363,045 chains-of-thought and 199,589 QA pairs. The resulting 7B OmniChem matches GPT 4o mini on custom and ChemBench, while reducing hallucinations by 79.46%, and supports molecular design and synthesis planning for scalable scientific knowledge discovery.","arXiv :2607 .08403v 1 [ cs .AI] 9 Jul 2026  \nGame Theory Driven Multi-Agent Framework Mitigates Language  \nModel Hallucination  \nRunzhe Liu 1 ,2 , Biquan Bie4 , Zihao Wang3 , Yuchao Ma 1 ,2 , Yexin Liu3 , Xinghai Li 1 ,2  \nHarry Yang3 , Wenbo Yang 1 ,2 ,∗ , Jinzhe Cao 1 ,2 , Shengyang Tao 1 ,2 ,∗  \n1 State Key Laboratory of Fine Chemicals, Frontier Science Center for Smart Materials, Dalian University of Technology, Dalian, 116024, China  \n2 Dalian Key Laboratory of Intelligent Chemistry, CR Belt and Road Joint Laboratory on Intelligent Chemistry and Advanced Materials of Liaoning Province, School of Chemistry, Dalian University of  \nTechnology, Dalian, 116024, China  \n3 Academy of Interdisciplinary Studies, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong  \n4 Independent Researcher, Beijing 100032, China.  \n∗ Correspondence: [wbyang@dlut.edu.cn](wbyang@dlut.edu.cn) ; [taosy@dlut.edu.cn](taosy@dlut.edu.cn)  \nAbstract  \nThe application of lightweight Large Language Models in rule-based scientific domains remains severely limited by their tendency to mimic linguistic patterns rather than reproduce axiomatic reasoning, causing frequent hallucinations. Here, we show that G-Frame, an adaptive multi-agent framework integrating Bayesian and team game principles, establishes an automated closed-loop for high-quality data synthesis and model training. By forcing the internalization of domain constraints through structured reasoning, we synthesized a specialized corpus of 363,045 chains-of-thought and 199,589 question-answer pairs. The resulting 7B model OmniChem achieves performance parity with GPT 4o mini on custom benchmarks and ChemBench while exhibiting a 79.46% reduction in hallucinations relative to its base architecture. We further demonstrate the advanced capabilities of OmniChem in molecular design and synthesis planning. This work establishes a scalable paradigm utilizing adaptive multi-agents to overcome inherent reasoning deficiencies, offering a feasible pathway for accelerating knowledge discovery in specialized scientific fields.  \n1 Introduction  \nTransformer-based Large Language Models (LLMs) show significant potential to transform scientific research in knowledge discovery and experimental design[1, 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10] . Lightweight LLMs offer a compelling alternative by providing secure and scalable local deployment options versus closed-source commercial APIs[11, 12 , 13 , 14 , 15] . Task-specific fine-tuning allows these smaller models to achieve performance parity with larger counterparts at significantly reduced computational costs.  \nApplying general-purpose LLMs to rigorous scientific fields like chemistry faces fundamental challenges. Their probabilistic autoregressive nature struggles to replicate structured expert reasoning or internalize axiomatic physical constraints. This limitation results in plausible yet factually incorrect text generation[1, 16 , 17] . Such hallucinations are particularly acute in lightweight models and obstruct their reliable application in drug discovery or materials design[18, 19 , 20 , 21 , 22] .  \nExisting approaches including tool integration or supervised fine-tuning have inherent limitations[23, 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35] . Reliance on closed-source APIs compromises data sovereignty while offline model fine-tuning is restricted by specialized data  \navailability. This study therefore addresses the generation of factually unsupported text by lightweight LLMs within specialized domains like chemistry. This issue stems from the failure of autoregressive mechanisms to reproduce structured causal reasoning, and its resolution is critical for enhancing model reliability and reasoning in complex scientific tasks.  \nMitigating factual inaccuracies arising from inadequate domain reasoning requires internalizing physical principles via complex training optimization. To address this challenge, we devel","cbCaisfMdlHrNBos","https://ap.wps.com/l/cbCaisfMdlHrNBos","pdf",4022688,3,1,34,"English","en",105,"# Introduction\n# G-Frame Game-Theoretic Multi-Agent Framework","[{\"question\":\"Why do lightweight large language models hallucinate in scientific domains like chemistry?\",\"answer\":\"Their probabilistic autoregressive generation tends to mimic linguistic patterns rather than reproduce structured expert reasoning or internalize axiomatic physical constraints. This leads to plausible but factually incorrect outputs, especially in lightweight models.\"},{\"question\":\"What is G-Frame and how does it reduce hallucinations?\",\"answer\":\"G-Frame is an adaptive multi-agent framework that integrates team game and Bayesian game principles into a hierarchical probabilistic optimization loop. It suppresses entropy at the micro level and dynamically optimizes decisions under uncertainty at the macro level, internalizing domain constraints through structured reasoning.\"},{\"question\":\"What results does OmniChem achieve after training with G-Frame?\",\"answer\":\"OmniChem (a 7B model) achieves performance parity with GPT 4o mini on custom benchmarks and ChemBench, and shows a 79.46% reduction in hallucinations versus its base architecture. It also demonstrates capabilities in molecular design and synthesis planning.\"}]",1784200938,86,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"game-theory-driven-multi-agent-framework-mitigates-language-model-hallucination","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/game-theory-driven-multi-agent-framework-mitigates-language-model-hallucination/85082/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-24","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},"Why do lightweight large language models hallucinate in scientific domains like chemistry?","Question",{"text":75,"@type":76},"Their probabilistic autoregressive generation tends to mimic linguistic patterns rather than reproduce structured expert reasoning or internalize axiomatic physical constraints. This leads to plausible but factually incorrect outputs, especially in lightweight models.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What is G-Frame and how does it reduce hallucinations?",{"text":80,"@type":76},"G-Frame is an adaptive multi-agent framework that integrates team game and Bayesian game principles into a hierarchical probabilistic optimization loop. It suppresses entropy at the micro level and dynamically optimizes decisions under uncertainty at the macro level, internalizing domain constraints through structured reasoning.",{"name":82,"@type":73,"acceptedAnswer":83},"What results does OmniChem achieve after training with G-Frame?",{"text":84,"@type":76},"OmniChem (a 7B model) achieves performance parity with GPT 4o mini on custom benchmarks and ChemBench, and shows a 79.46% reduction in hallucinations versus its base architecture. It also demonstrates capabilities in molecular design and synthesis planning.","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":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]