[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-657630-105":59,"doc-detail-657630-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","diagrameval-evaluating-llm-generated-diagrams-via-graphs","DiagramEval - Evaluating LLM-Generated Diagrams via Graphs","","Diagrams are crucial in research writing, yet they are difficult and expensive to create manually, and conventional image generation models often fail to produce diagrams with clear, well-structured layouts. The paper studies evaluating LLM-generated demonstration diagrams created as SVG text. It introduces DiagramEval, modeling diagrams as directed graphs and using two metric groups—node alignment and path alignment—to measure quality, validate against state-of-the-art systems, and provide stronger explainability.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/diagrameval-evaluating-llm-generated-diagrams-via-graphs/657630/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/diagrameval-evaluating-llm-generated-diagrams-via-graphs/657630.png","ImageObject",300,407,{"name":92,"@type":93},"lugxt","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-30","2026-10-26",true,{"@type":102,"interactionType":103,"userInteractionCount":81},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"Why are existing diagram generation approaches hard to evaluate reliably?","Question",{"text":112,"@type":113},"Existing evaluation metrics often rely on general vision, which limits explainability and its ability to capture fine-grained logical correctness, leading to only moderate agreement with human judgments.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does DiagramEval represent a diagram for evaluation?",{"text":117,"@type":113},"DiagramEval converts diagrams into SVG, parses text elements as nodes, and extracts connections as directed edges to build a reference and generated graph.",{"name":119,"@type":110,"acceptedAnswer":120},"What metric groups does DiagramEval use?",{"text":121,"@type":113},"It uses two new groups of metrics: node alignment and path alignment, quantified with precision, recall, and F1 scores over nodes and multi-hop edges.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},657630,1793136758,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":81,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145},3985747862600,"https://ap-avatar.wpscdn.com/davatar_9964176cb1d06d4a9deccf72a44ae3dc","DiagramEval: Evaluating LLM-Generated Diagrams via Graphs  \nChumeng Liang1 , Jiaxuan You1  \n1University of Illinois Urbana-Champaign  \n[chumengl@illinois.edu](chumengl@illinois.edu), [jiaxuan@illinois.edu](jiaxuan@illinois.edu)  \narXiv :2510 .25761v2 [ cs .CL] 31 Oct 2025  \nAbstract  \nDiagrams play a central role in research papers for conveying ideas, yet they are often notoriously complex and labor-intensive to create. Although diagrams are presented as images, standard image generative models struggle to produce clear diagrams with well-defined structure.  \nWe argue that a promising direction is to generate demonstration diagrams directly in textual form as SVGs, which can leverage recent advances in large language models (LLMs) .  \nHowever, due to the complexity of components and the multimodal nature of diagrams, sufficiently discriminative and explainable metrics for evaluating the quality of LLM-generated diagrams remain lacking. In this paper, we propose DiagramEval, a novel evaluation metric designed to assess demonstration diagrams generated by LLMs. Specifically, DiagramEval conceptualizes diagrams as graphs, treating text elements as nodes and their connections as directed edges, and evaluates diagram quality using two new groups of metrics: node alignment and path alignment. For the first time, we effectively evaluate diagrams produced by stateof-the-art LLMs on recent research literature, quantitatively demonstrating the validity of our metrics. Furthermore, we show how the enhanced explainability of our proposed metrics offers valuable insights into the characteristics  \nof LLM-generated diagrams. Code: [https:](https:)//[github.com/ulab-uiuc/diagram-eval](github.com/ulab-uiuc/diagram-eval).  \n1 Introduction  \nDiagrams play a central role in research papers for conveying ideas, e.g., the diagram of Transformer (Vaswani et al., 2017) has played a pivotal role in presenting and publicizing the idea, making it one of the most cited deep learning papers. However, generating high-quality diagrams is often notoriously complex and labor-intensive to create. Consequently, automated diagram generation is a central challenge in AI-assisted scientific discovery (Eger et al., 2025), potentially saving millions  \nof hours for researchers while improving the quality of research publications. Although diagrams are presented as images, standard image generative models struggle to produce clear diagrams with well-defined structure (Zala et al., 2023) . We argue that a promising direction is to generate demonstration diagrams directly in textual form as SVGs, which can leverage recent advances in large language models (LLMs) .  \nExisting methods on automated diagram generation with LLMs rely heavily on proprietary LLMs, either through direct planning (Mondal et al., 2024 ; Zhang et al., 2024 ; Cui et al., 2025) or as assistance in diagram generation (Belouadi et al., 2023, 2024 ; Cui et al., 2025) . Given this reliance, advancements in cutting-edge proprietary LLMs, such as Claude 3.7 Sonnet (Anthropic, 2025) and Gemini 2.5 Pro (Google DeepMind, 2025), directly enhance automated diagram generation through SVG code generation capabilities (Blecher et al., 2023) .  \nHowever, existing evaluation metrics lack the expressiveness needed to differentiate diagrams of varying quality. Most benchmarks employ modelbased, diagram-level metrics (Hessel et al., 2021 ; Fu et al., 2023), which were originally designed for general text-to-image tasks rather than textwithin-image generation specific to diagrams (Rodriguez et al., 2023) . These metrics evaluate entire diagrams using general vision models, which limits their explainability and their ability to accurately capture detailed logical correctness, resulting in only moderate correlation with human judgments (Eger et al., 2025) . Thus, there is a critical need for new metrics.  \nIn this paper, we propose DiagramEval, a novel evaluation metric designed to assess demonstration diagrams genera","cbCaivlOs4zlRpWR","https://ap.wps.com/l/cbCaivlOs4zlRpWR","pdf",1891480,13,"English","# Abstract\n# 1 Introduction","[{\"question\":\"Why are existing diagram generation approaches hard to evaluate reliably?\",\"answer\":\"Existing evaluation metrics often rely on general vision, which limits explainability and its ability to capture fine-grained logical correctness, leading to only moderate agreement with human judgments.\"},{\"question\":\"How does DiagramEval represent a diagram for evaluation?\",\"answer\":\"DiagramEval converts diagrams into SVG, parses text elements as nodes, and extracts connections as directed edges to build a reference and generated graph.\"},{\"question\":\"What metric groups does DiagramEval use?\",\"answer\":\"It uses two new groups of metrics: node alignment and path alignment, quantified with precision, recall, and F1 scores over nodes and multi-hop edges.\"}]","DiagramEval - Evaluating LLM-Generated Diagrams via Graphs | PDF",1792985678,33]