[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83851-en":3,"doc-seo-83851-105":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},83851,8796095462418,"Noah","https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780",8,"Research & Report","Dashboard2Code: Evaluating Multimodal Models on Reconstructing Interactive Dashboards","Automatic data visualization generation has advanced with multimodal large language models, yet most work targets static charts and ignores interactive dashboards used for real data exploration. Dashboard2Code is introduced as a proactive task: explore a dashboard, collect feedback from interactions such as clicking and filtering, and generate code that reproduces both the target visuals and interaction logic. DashboardMimic provides 180 verified Plotly+Dash dashboard–code pairs across three difficulty levels and eight interaction patterns. An evaluation framework blends code semantic analysis with dynamic interaction-based testing, aligning strongly with human judgments. Experiments show high-complexity limitations and a persistent open- vs closed-source gap.","Dashboard2Code: Evaluating Multimodal Models on Reconstructing  \nInteractive Dashboards  \nTianhao Niu* Ziyu Han* Qiguang Chen  \nShiqi Zhou Baocai Shan Hengjie Fang Qingfu Zhu Wanxiang Che†  \nResearch Center for Social Computing and Interactive Robotics Harbin Institute of Technology, China  \n{thniu,zyhan,[car}@ir.hit.edu.cn](car}@ir.hit.edu.cn)  \narXiv :2607 .04727v 1 [ cs . SE] 6 Jul 2026  \nAbstract  \nAutomatic data visualization generation has advanced rapidly with multi-modal large language models, yet existing efforts largely focus on static charts and overlook the interactive dashboards commonly used for real-world data exploration. We introduce Dashboard2Code, a novel task that requires a model to proactively explore an interactive dashboard, acquire and integrate feedback from its own interactions (e.g., clicking and filtering), and generate code that reproduces the target dashboard. To support comprehensive evaluation, we present DashboardMimic, the first Plotly+Dash benchmark for Dashboard2Code, comprising 180 carefully designed and manually verified dashboard–code pairs spanning three difficulty levels and covering eight common real-world interaction patterns. We further propose an automated evaluation framework tailored to dashboards that combines code semantic analysis with dynamic interaction-based testing to assess visual and interaction consistency, showing strong agreement with human judgments. Experiments across a range of open-and closedsource multi-modal models reveal that even the strongest systems struggle on high-complexity dashboards and that a substantial performance gap remains between open-source and closedsource models on the Dashboard2Code task. 1  \n1 Introduction  \nAutomatic data visualization understanding and generation are important tasks in artificial intelligence. With the rapid development of multi-modal large language models, tasks for understanding data visualizations (e.g., chart question answering (Masry et al., 2022, 2025), chart summarization (Choi et al., 2025 ; Wang et al., 2024), chart parsing (Xu et al., 2025b)) and tasks for generating  \n*Equal contribution.†Corresponding author.  \n1 Code and data  \n(a) Chart2Code  \n(b) Video2Code  \nFigure 1: Comparison of Dashboard2Code task with existing tasks. Dashboard2Code requires the model to proactively explore a dashboard, collect feedback during exploration, and integrate this feedback with the interaction history to generate code that faithfully reproduces both the visual appearance and interaction logic of the target dashboard.  \nvisualizations (e.g., text-to-chart (Rahman et al., 2025 ; Ni et al., 2025) and chart-to-code (Yang et al., 2025 ; Wu et al., 2025 ; Zhao et al., 2025 ; Tang et al., 2025a)) show strong results. However, these tasks mainly focus on understanding and generating static visualizations. In practical scenarios, datasets are often large and structurally complex, making interactive dashboards a common form of visualization for in-depth exploration. As interactive interfaces, dashboards enable users to uncover more complex data patterns and logical relations through actions such as clicking and filtering.  \nWith the rapid development of multi-modal large language models (Wang et al., 2025a ; OpenAI et al., 2024 ; Team et al., 2025), several important research  \ntasks have emerged on automating interactive user interfaces.  \nOne line of research systematically evaluates multi-modal GUI agents on perception, action, and reasoning over interactive interfaces. For instance, OS-World (Xie et al., 2024) introduces a benchmark that evaluates multi-modal agents in real operating system environments on open-ended computer tasks involving diverse applications, interfaces, and workflows. In the data visualization domain, DashboardQA (Kartha et al., 2025) presents the dashboard question answering benchmark that assesses a GUI agent’s ability to interact with a dashboard in response to a given question and return the correct final ","cbCaioTNX6CQPNdB","https://ap.wps.com/l/cbCaioTNX6CQPNdB","pdf",6274914,1,37,"English","en",105,"# Abstract\n# Introduction\n# Code and data\n## Chart2Code\n## Video2Code","[{\"question\":\"What is the Dashboard2Code task and how does it differ from prior visualization code-generation tasks?\",\"answer\":\"Dashboard2Code requires the model to proactively explore an interactive dashboard, collect feedback during actions like clicking and filtering, and integrate that feedback with its interaction history to generate code that reproduces both visuals and interaction logic. Prior work often focuses on static snapshots or relies on human-annotated interaction traces, making the model more passive.\"},{\"question\":\"What benchmark dataset is proposed for Dashboard2Code evaluation?\",\"answer\":\"DashboardMimic is introduced as the first Plotly+Dash benchmark for Dashboard2Code, containing 180 carefully designed and manually verified dashboard–code pairs across three difficulty levels. It also covers eight common real-world interaction patterns.\"},{\"question\":\"How is model performance evaluated for dashboards in this work?\",\"answer\":\"The paper proposes an automated evaluation framework that combines code semantic analysis with dynamic interaction-based testing. This assesses both visual and interaction consistency and shows strong agreement with human judgments.\"}]",1784190975,93,{"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},"dashboard2code-evaluating-multimodal-models-on-reconstructing-interactive-dashboards","",{"@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/dashboard2code-evaluating-multimodal-models-on-reconstructing-interactive-dashboards/83851/",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 is the Dashboard2Code task and how does it differ from prior visualization code-generation tasks?","Question",{"text":75,"@type":76},"Dashboard2Code requires the model to proactively explore an interactive dashboard, collect feedback during actions like clicking and filtering, and integrate that feedback with its interaction history to generate code that reproduces both visuals and interaction logic. Prior work often focuses on static snapshots or relies on human-annotated interaction traces, making the model more passive.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What benchmark dataset is proposed for Dashboard2Code evaluation?",{"text":80,"@type":76},"DashboardMimic is introduced as the first Plotly+Dash benchmark for Dashboard2Code, containing 180 carefully designed and manually verified dashboard–code pairs across three difficulty levels. It also covers eight common real-world interaction patterns.",{"name":82,"@type":73,"acceptedAnswer":83},"How is model performance evaluated for dashboards in this work?",{"text":84,"@type":76},"The paper proposes an automated evaluation framework that combines code semantic analysis with dynamic interaction-based testing. This assesses both visual and interaction consistency and shows strong agreement with human judgments.","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"]