智能数据洞察：大模型驱动的可视化与可视分析

如果无法正常显示，请先停止浏览器的去广告插件。

1. 人智融合数据洞察：大模型驱动的智能可视分析与故事叙述复旦大学大数据学院陈思明青年研究员，博士生导师 2025年11月

2. 大数据可视化与可视分析 - 复旦大学大数据学院，陈思明本科：复旦大学计算机学院博士：北京大学智能科学与技术博士后：德国波恩大学研究科学家：德国Fraunhofer IAIS 青年研究员/博士生导师：复旦大学大数据学院扬帆计划获得者、上海市高层次引进人才。发表论文100余篇，在 IEEE Visualization, IEEE TVCG, ACM CHI等重要国际顶级可视化会议以及期刊上发表40余篇文章（CCF A）。被评为AI2000可视化领域十年间最有影响力学者提名奖（全球100）。担任3个期刊（2个SCI期刊）的副主编与编委、多个国际会议的组织委员会和程序委员会成员，包括IEEE PacificVis论文（短文）主席，ChinaVis论文主席、数据分析挑战赛主席。他的工作曾获得多次IEEE VAST Challenge数据挑战赛一等奖，以及多个会议最佳论文/海报（提名）奖。

3. 人智融合数据洞察大模型驱动智能可视分析数据洞察的高效信息传递多模态的创意信息增强复杂决策与交互分析文本+数据：构建数据故事视频视频+可视化：融合真实与虚拟人智协同的深度洞察挖掘可视化+数据：智能数据总结3D特效场景交互式创作 AR+VIS探索

4. 4 !"#$ 1. 可视分析系统制作费时费力 -- 至少需要200人时。 2. 只支持固定数据的分析 -- 无法响应用户实时的需求。解决方案：自动化生成可视分析大屏来解决用户的数据分析需求。 traditional visual analytics system[1] 2 [1] T. Qiu et al., "FishEye Watcher: A Visual Analytics System for Knowledge Graph Bias Detection - IEEE VAST Challenge 2024 MC1 Honorable Mention for Effective Use of Coordinated Views to Interrogate Bias," 2024 IEEE Visual Analytics Science and Technology VAST Challenge, St. Pete Beach, FL, USA, 2024, pp. 13-14, doi: 10.1109 FDUVIS 4

5. 5 %& } 挑战 } 1. 缺少可视分析知识 } 2. 缺少联系与交互 • 动机: 通过人机协作方式，自动化的生成拥有可视化知识的、可交互的关联视图。 5 FDUVIS 5

6. 6 SmartMLVs '( 7 FDUVIS 6

7. 7 )*+, FDUVIS 7 1 4

8. 8 FDUVIS 8

9. 1 | 研究背景和相关工作 -./01234)56789:;<=>?@ABC LLM-Agent assistantce 分析人员可视分析系统 C1. 需要根据分析目标分解任务C1. 系统中往往缺乏探索中的指导 C2. 随着探索的进行，分析需求灵活多变C2. 系统是预设的，非自适应于任务 C3. 缺乏可视分析的实现技能C3. 要求开发人员编写代码 FDUVIS 9

10. 2 | 研究框架和内容 DE'( !"#$%&'()*+,-./0123456789 • 基础：分析目标、数据 • 经过： • 任务规划（推荐与分解） • 任务执行（代码生成） • 结果：洞察 • 合作模式：模型负责规划与执行，人负责监督和完善 FDUVIS 10

11. 2 | 研究框架和内容 DE'( !"#$%&'():;23<1=>?@1*2,-14567 1. 任务推荐：把目标转换成可执行的任务 2. 任务执行：生成解决任务所需的可视化和数据建模的代码，由建模结果或者人的探索获得洞察。单个可视化可合并成多个可视化组成的系统。相应的，洞察也可自由合并进行总结。 3. 任务分解：评估任务执行结果，若解决不充分则进一步分解成子任务。 FDUVIS 11

12. 2 | 研究框架和内容研究框架任务的规划与执行基于任务流：task flow 任务（节点）任务推荐（边）任务分解（边）任务推荐（初始推荐和有历史记录推荐）初始推荐表示新提出的任务 FDUVIS 已有分析结果后推荐新任务任务分解（并行/串行） and（并行） down（串行） 12

13. 2 | 研究框架和内容 LightVA =>A5BCDEFG5B 用于选择要合并的可视化查看分析进度回顾分解过程颜色表示已探索的数据空间 FDUVIS 13

14. 2 | 研究框架和内容使用场景：社交媒体事件分析 FDUVIS 14

15. ProactiveVA: 主动式UI智能体驱动的可视分析 User VA system UI Agent Proac&veVA: Proac&ve Visual Analy&cs with LLM-based UI Agent FDUVIS 15

16. ProactiveVA: 主动式UI智能体驱动的可视分析 Visual Analytics Interface LightVA 自动化生成 ProactiveVA 数据算法感知可视化交互主动式感知洞察可视分析的愿景和目标是实现人与ai的智能协同分析和理解数据。 FDUVIS 16

17. 三种可视分析的范式比较传统可视分析 FDUVIS 响应式可视分析主动式可视分析 17

18. 主动式可视分析研究挑战 AI Agent 现有工作时间和解决方案依赖于预定义的规则 FDUVIS Proactively assist User Timing 时机 When to helpC1 How to helpC2 Solution 方法我们的工作自主决定何时以及如何协助 18

19. ProactiveVA的工作流程：“感知-推理-行动”

20. ProactiveVA的工作流程：“感知-推理-行动” Visual analytics interface 操作系统界面交互来发现洞察观察用户的交互行为、理解意图 FDUVIS 20

21. 主动式UI智能体辅助的可视分析系统界面 Chat View Notes View VA Interface View

22. Proactive UI Agent-assisted VA Interface 智能体操作系统并高亮发现的数据洞察 Chat View VA Interface View Notes View 用户操作界面智能体解释操作过程用户提交笔记智能体检查用户的笔记

23. ProactiveVA的两个用例 🏆 2021 IEEE 可视分析挑战赛的获奖作品 Data: 社交媒体数据 Task: 识别事件并评估阿比拉市的风险 📈 一个经典的销售和利润分析表仪表板 Data: 销售数据 Task: 分析销售和利润

24. 系统上手过程的辅助 - demo

25. 数据探索过程的辅助 – vast challenge - demo

26. 数据探索过程的辅助 – tableau sales dashboard -demo

27. 人智融合数据洞察大模型驱动智能可视分析数据洞察的高效信息传递多模态的创意信息增强复杂决策与交互分析文本+数据：构建数据故事视频视频+可视化：融合真实与虚拟人智协同的深度洞察挖掘可视化+数据：智能数据总结3D特效场景交互式创作 AR+VIS探索 FDUVIS 27

28. Narrative Player：让数据叙述增加动态可视化效果 Narrative Player: Reviving Data Narratives with Visuals FDUVIS 28

29. 数据文档 FDUVIS 29

30. 数据视频 FDUVIS 30

31. 数据叙事文本 1 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. 2 Did you know, 3 the hottest months are from May to August? 4 It’s a symphony of heat, 5 where daily mean temperatures often flirt with the mercury‘s upper 30℃. But the city doesn’t just tip its hat to the sun. 6 Rainfall plays its part, 7 especially in May and June, 8 where it reaches its crescendo with over 250mm, 9 making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. 10 The chill of winter finds its way, 10’ 11’ 11 with December‘s average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. 12 The rain seems to sense the need for tranquility, 13 with precipitation also taking a back seat in winter. FDUVIS 数据表格 Month Jan Feb Mar Apr May … Record High Average High Daily Mean 28.3 18.2 14.2 31.4 20.4 16.4 32.4 22.3 18.6 34.4 26.3 22.6 34.9 30.0 26.0 … … … 31 … … … … … … …

32. 数据叙事文本数据表格 1 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. 2 Did you know, 3 the hottest months are from May to August? 4 It’s a symphony of heat, 5 where daily mean temperatures often flirt with the mercury‘s upper 30℃. But the city doesn’t just tip its hat to the sun. 6 Rainfall plays its part, 7 especially in May and June, 8 where it reaches its crescendo with over 250mm, 9 making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. 10 The chill of winter finds its way, 10’ 11’ 11 with December‘s average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. 12 The rain seems to sense the need for tranquility, 13 with precipitation also taking a back seat in winter. 数据视频 1 2 FDUVIS 3 4 5 6 7 Narrative Player 9 8 Month Jan Feb Mar Apr May … 10 Record High Average High Daily Mean 28.3 18.2 14.2 31.4 20.4 16.4 32.4 22.3 18.6 34.4 26.3 22.6 34.9 30.0 26.0 … … … 11 12 13 32 … … … … … … …

33. 数据叙事文本数据表格 1 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. 2 Did you know, 3 the hottest months are from May to August? 4 It’s a symphony of heat, 5 where daily mean temperatures often flirt with the mercury‘s upper 30℃. But the city doesn’t just tip its hat to the sun. 6 Rainfall plays its part, 7 especially in May and June, 8 where it reaches its crescendo with over 250mm, 9 making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. 10 The chill of winter finds its way, 10’ 11’ 11 with December‘s average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. 12 The rain seems to sense the need for tranquility, 13 with precipitation also taking a back seat in winter. 数据视频 1 2 FDUVIS 3 4 5 6 7 Narrative Player 9 8 Month Jan Feb Mar Apr May … 10 Record High Average High Daily Mean 28.3 18.2 14.2 31.4 20.4 16.4 32.4 22.3 18.6 34.4 26.3 22.6 34.9 30.0 26.0 … … … 11 12 13 33 … … … … … … …

34. Narrative Player — — Reviving Data Narratives with Visuals 叙事文本叙事分析可视化生成文本切片与分类基于数据事实的可视化映射子句数据事实集提取与验证初始数据事实集数据表格 FDUVIS 数据事实集推理与补全数据事实候选集可视化候选集可视化序列选择数据视频选中的可视化序列多通道效果增强 34

35. 工作流 — 叙事分析 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. Did you know, the hottest months are from May to August? It's a symphony of heat, where daily mean temperatures often flirt with the mercury's upper 30℃. 叙事文本 But the city doesn't just tip its hat to the sun. Rainfall plays its part, especially in May and June, where it reaches its crescendo with over 250mm, making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. The chill of winter finds its way, with December's average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. The rain seems to sense the need for tranquility, with precipitation also taking a back seat in winter. 数据表格 FDUVIS Month Jan Feb Mar Apr May … Record High Average High Daily Mean 28.3 18.2 14.2 31.4 20.4 16.4 32.4 22.3 18.6 34.4 26.3 22.6 34.9 30.0 26.0 … … … … … … … … … … 35

36. 工作流 — 叙事分析 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. Did you know, the hottest months are from May to August? It's a symphony of heat, where daily mean temperatures often flirt with the mercury's upper 30℃. But the city doesn't just tip its hat to the sun. 子句 Rainfall plays its part, especially in May and June, where it reaches its crescendo with over 250mm, making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. The chill of winter finds its way, with December's average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. The rain seems to sense the need for tranquility, FDUVIS with precipitation also taking a back seat in winter. 36

37. 工作流 — 叙事分析 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. Did you know, the hottest months are from May to August? It's a symphony of heat, where daily mean temperatures often flirt with the mercury's upper 30℃. But the city doesn't just tip its hat to the sun. 子句 Rainfall plays its part, especially in May and June, where it reaches its crescendo with over 250mm, making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. The chill of winter finds its way, with December's average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. The rain seems to sense the need for tranquility, FDUVIS with precipitation also taking a back seat in winter. 37

38. 工作流 — 叙事分析目标子句 where daily mean temperatures often flirt with the mercury's upper 30℃. C 初步抽取出的数据事实 "Type": "correlate", F1 "Parameter": null, "Measures": ["Daily mean", "Record high"], "Context": {}, "Breakdowns": ["Month"], "Focus": ["May", "Jun", "Jul", "Aug"] "Type": "distribution", F2 "Parameter": null, "Measures": ["Daily mean"], "Context": {}, "Breakdowns": ["Month"], "Focus": ["May", "Jun", "Jul", "Aug"] "Type": "deviation", F3 "Parameter": 30, "Measures": ["Daily mean"], "Context": {}, "Breakdowns": ["Month"], "Focus": ["May", "Jun", "Jul", "Aug"] FDUVIS 数据事实集提取与验证数据事实抽取的提示词 You are a narration analyst who specializes in converting clauses of long narratives individually into multiple facts. {{Definition of fact format and individual elements}} {{Few examples to guide the generation}} You are required to do the extraction. The data is [data table]. The narration is [narration]. The ClauseList is [clause1, clause2, …]. List 3 possible fact candidates in JSON as [clause1: {F1, F2, F3}, .... ], where F1, F2, F3 are 3 fact candidates. 重写数据事实的提示词 The fact candidate is : {F1, F2, F3}, The data table is [data table], The whole context is: "…, the hottest months are from May to August? It‘s a symphony of heat, {} , …” Write Clauses {C1, C2, C3} individually based on {F1, F2, F3} to fill in the blank of the context. 重写后的子句 where the daily mean temperature and the record high show a strong positive correlation. 𝑆𝑖𝑚(𝐸! , 𝐸! ) = 0.804 # C 1 with the daily mean temperature soaring above 26°C for each C of these months. 𝑆𝑖𝑚(𝐸!" , 𝐸! ) = 2 0.841 as the daily mean temperature deviates significantly from the C 30°C mark. 3 𝑆𝑖𝑚(𝐸!! , 𝐸! ) = 0.903 38

39. 工作流 — 叙事分析 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. Did you know, the hottest months are from May to August? It's a symphony of heat, where daily mean temperatures often flirt with the mercury's upper 30℃. But the city doesn't just tip its hat to the sun. 子句 Rainfall plays its part, especially in May and June, where it reaches its crescendo with over 250mm, making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. The chill of winter finds its way, with December's average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. The rain seems to sense the need for tranquility, FDUVIS with precipitation also taking a back seat in winter. 39

40. 工作流 — 叙事分析 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. Did you know, the hottest months are from May to August? It's a symphony of heat, where daily mean temperatures often flirt with the mercury's upper 30℃. But the city doesn't just tip its hat to the sun. 子句 Rainfall plays its part, especially in May and June, where it reaches its crescendo with over 250mm, making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. The chill of winter finds its way, 模糊子句 with December's average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. The rain seems to sense the need for tranquility, FDUVIS with precipitation also taking a back seat in winter. 40

41. 工作流 — 叙事分析 … where daily mean temperatures oden ﬂirt with the mercury's upper 30s… The chill of winter finds its way, 模糊子句 with December‘s average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. Properties: Daily mean Values：May, Jun, Jul, Aug Properties: Daily mean, Average low, Record low, Average high, Record high “chill” Values: Dec, Jan, Feb “winter” Properties: Average low, Record low Values：Dec FDUVIS 数据事实集推理与补全推断出的数据事实 … "Measures": ["Daily mean"], "Focus": ["Dec"], …F1 … "Measures": ["Average low", "Record low", "Daily mean"], "Context": {"Month": ["Dec"]}, …F2 … "Measures": ["Average low", "Record low"], "Context": {"Month": ["Dec"]}, …F3 41

42. 工作流 — 叙事分析 The chill of winter finds its way, 数据事实集提取与验证数据事实集推理与补全 … … 数据事实 FDUVIS 42

43. 工作流 — 可视化生成 The chill of winter finds its way, FDUVIS …… …… 43

44. 工作流 — 可视化生成最小化过渡损失强调视觉聚焦优化激活“主要可视化” 子句 i+1 可视化备选集 i+1 可视化备选集 i 子句 i The chill of winter finds its way, 子句 i-1 FDUVIS … … 可视化备选集i-1 44

45. 工作流 — 可视化生成 1 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. 2 Did you know, 3 the hottest months are from May to August? 4 It’s a symphony of heat, 5 where daily mean temperatures often flirt with the mercury‘s upper 30℃. But the city doesn’t just tip its hat to the sun. 6 Rainfall plays its part, 7 especially in May and June, 8 where it reaches its crescendo with over 250mm, 9 making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. 10 The chill of winter finds its way, 10’ 11’ 11 with December‘s average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. 12 The rain seems to sense the need for tranquility, 13 with precipitation also taking a back seat in winter. 1 2 FDUVIS 3 4 5 6 7 8 9 10 11 12 13 45

46. 工作流 — 可视化生成 1 Nestled in the heart of Guangdong, Zhaoqing’s climate tells a story as rich as its heritage. 2 Did you know, 3 the hottest months are from May to August? 4 It’s a symphony of heat, 5 where daily mean temperatures often flirt with the mercury‘s upper 30℃. But the city doesn’t just tip its hat to the sun. 6 Rainfall plays its part, 7 especially in May and June, 8 where it reaches its crescendo with over 250mm, 9 making it a weather sonata of heat and rain. But like every masterpiece, Zhaoqing has its quieter notes too. 10 The chill of winter finds its way, 10’ 11’ 11 with December‘s average low at a cool 12.5°C and a record low that dips to an almost freezing 1.7°C. 12 The rain seems to sense the need for tranquility, 13 with precipitation also taking a back seat in winter. 1 2 3 4 5 6 7 8 9 10 11 12 13 One-to-Two transition Transition with intermediate frame FDUVIS 46

47. Example Case: The weather in Zhaoqing FDUVIS 47

48. FDUVIS 48

49. 人智融合数据洞察大模型驱动智能可视分析数据洞察的高效信息传递多模态的创意信息增强复杂决策与交互分析文本+数据：构建数据故事视频视频+可视化：融合真实与虚拟人智协同的深度洞察挖掘可视化+数据：智能数据总结3D特效场景交互式创作 AR+VIS探索 FDUVIS 49

50. !" !"#$%&'()*+,-!"./01234560789:;<=>?,-@AB CCDEF?GH./IJ Chart-to-Text1VisText2 DataTales3VL2NL KLMNOP 4 [1] Jason Obeid and Enamul Hoque. 2020. Chart-to-Text: Generating Natural Language Descriptions for Charts by Adapting the Transformer Model. In Proceedings of the 13th International Conference on Natural Language Generation, pages 138–147, Dublin, Ireland. Association for Computational Linguistics. [2] Benny Tang, Angie Boggust, and Arvind Satyanarayan. 2023. VisText: A Benchmark for Semantically Rich Chart Captioning. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7268–7298, Toronto, Canada. Association for Computational Linguistics. [3] Sultanum N, Srinivasan A. Datatales: Investigating the use of large language models for authoring data-driven articles[C]//2023 IEEE Visualization and Visual Analytics (VIS). IEEE, 2023: 231-235. [4] KoFDUVIS H K, Jeon H, Park G, et al. Natural language dataset generation framework for visualizations powered by large language models[C]//Proceedings of the 2024 CHI Conference on Human Factors in Computing50 Systems. 5 0

51. #$ 5 1 FDUVIS 51

52. #$ !"#$ 3 FDUVIS 52

53. #$ !"#$ %&'( 3 %&'( FDUVIS 53

54. #$ !"#$ %&'( 3 %&'( FDUVIS )"#$ 54

55. %&!"'() CCDQHRSEFTKMUVWX!"#$?YZ 4 *+,-./01)234 '56789:;<)=>?@ABC FDUVIS DEFGHIJKLMNO PQ8RS 55

56. %&!"'() CCDQHRSEFTKMUVWX!"#$?YZ `abcdefg)2KLMNhi 5 *+,-./ '56789:;< FDUVIS 01)234 DEFGHIJKLMNO PQ8RS )=>?@ABC TUV8KLMN8WXYZ[\]^_ 56

57. *+,- [' C\]C^! _`0_abcde!"#$?fH./ 6 [1] A. Lundgard and A. Satyanarayan. Accessible visualization via natural language descriptions: A four-level model of semantic content. IEEE Transactions on Visualization and Computer Graphics, 28(1):1073–1083, 2022. FDUVIS 57

58. *+,- [' C\]C^! _`0_abcde!"#$?fH./ Level 1: jklmnnopqrpst 6 [1] A. Lundgard and A. Satyanarayan. Accessible visualization via natural language descriptions: A four-level model of semantic content. IEEE Transactions on Visualization and Computer Graphics, 28(1):1073–1083, 2022. FDUVIS 58

59. *+,- [' C\]C^! _`0_abcde!"#$?fH./ Level 1: jklmnnopqrpst Level 2: 统计特征nn峰值、分布等 6 [1] A. Lundgard and A. Satyanarayan. Accessible visualization via natural language descriptions: A four-level model of semantic content. IEEE Transactions on Visualization and Computer Graphics, 28(1):1073–1083, 2022. FDUVIS 59

60. *+,- [' C\]C^! _`0_abcde!"#$?fH./ Level 1: jklmnnopqrpst Level 3: uvwxvyznn)2{|t Level 2: 统计特征nn峰值、分布等 6 [1] A. Lundgard and A. Satyanarayan. Accessible visualization via natural language descriptions: A four-level model of semantic content. IEEE Transactions on Visualization and Computer Graphics, 28(1):1073– 1083, 2022. FDUVIS 60

61. *+,- ~•CCDwEFbcde!"#$bNY€<?gh`k 数据集独立标注幻觉类型合并结果最终检查与标注 🔍 gh_a 8 • \ijgh`k • \llmghno • pqCrsC^tuvCCDwEF! "#$x?fHyz{|}_` FDUVIS 61

62. !" 9 FDUVIS 62

63. !" !"#$%&'()*+ •bcde=>‚_ƒ„…?†c ‹Œ•Ž••‘†c 10 ‡ˆ=>?34‰Š FDUVIS 63

64. !" 11 !",-$%&'*+) .-/0123 FDUVIS 64

65. !" 4567)89:; 12 FDUVIS 65

66. !" <=>?@89:; 13 FDUVIS 66

67. !" 14 FDUVIS 67

68. Benchmark ’“”•–—˜™š›œ0•'žŸ ¡¢kw!"#$EFx?"< 15 FDUVIS 68

70. #$ 📈 18 ¡75 m Benchmark !" ✏ 34 ¡‰Š • 9°¢k±ChartInsighter vs. Gold / GPT-4/ VL2NL • ‰Š±²‘ + ³´µKL{¶·¸{¶¹º»¼ • UVWX» =µL2 + L3 ®¯=¼/ ®¯~= • gh½ = gh®¯= / ®¯~= FDUVIS 70

71. ChartInsighter ChartInsighter0¾j•'¿-bcde!"#$EFxghÀÁ?ÂÃ0Œº%[K=>Ä ••'²‘EFbcde!"#$?ÅÆ •Ç“*ÈÉ?!"#$Benchmark 19 https://github.com/wangfen01/ChartInsig hter FDUVIS 71

72. 人智融合数据洞察大模型驱动智能可视分析数据洞察的高效信息传递多模态的创意信息增强复杂决策与交互分析文本+数据：构建数据故事视频视频+可视化：融合真实与虚拟人智协同的深度洞察挖掘可视化+数据：智能数据总结3D特效场景交互式创作 AR+VIS探索 FDUVIS 72

73. KinemaFX: 一种交互式粒子特效探索与创作系统 FDUVIS 73

74. 系统流程 a. 初始意图输入：用户结合语义文本(a1)和图形化输入(a2)来直观地表达设计意图 b. 粒子特效探索：通过在语义特征和运动特征权重控制(b1)与用户隐式偏好引导的迭代过程(b2)来探索粒子特效 c. 粒子特效组合：选择满意的粒子特效(c1)，调整并组合(c2)粒子特效形成完整的特效作品(c3) FDUVIS 74

75. 系统演示 FDUVIS 75

76. 用户实验效果展示 FDUVIS 76

77. 77 FDUVIS 77

78. FDUVIS 78

79. 相关文章 }Tian Qiu, Fen Wang, Shaohua Huang, Meng Guo, Yuheng Zhao, Jincheng Li, Siming Chen*. SmartMLVs: LLM-Enabled Multiple Linked Views Generation for Interactive Visualization. In Proceedings of IEEE PacificVis 2025, Best Paper Honorable Mention. }Yuheng Zhao, Junjie Wang, Linbin Xiang, Xiaowen Zhang, Zifei Guo, Cagatay Turkay, Yu Zhang, Siming Chen*. LightVA: Lightweight visual analytics with LLM agent-based task planning and execution. IEEE Transactions on Visualization and Computer Graphics, Accepted, 2024. }Zekai Shao, Leixian Shen, Haotian Li, Yi Shan, Huamin Qu, Yun Wang*, Siming Chen*. Narrative Player: Reviving Data Narratives with Visuals. IEEE Transactions on Visualization and Computer Graphics, Accepted, 2025. }Fen Wang, Bomiao Wang, Xueli Shu, Zhen Liu, Zekai Shao, Chao Liu, Siming Chen*. ChartInsighter: An Approach for Mitigating Hallucination in Time-series Chart Summary Generation with A Benchmark Dataset. IEEE Transactions on Visualization and Computer Graphics (IEEE PacificVis'25), Accepted, 2025. }Lin Gao, Leixian Shen, Yuheng Zhao, Jiexiang Lan, Huamin Qu, Siming Chen*. SceneLoom: Communicating Data with Scene Context. IEEE Transactions on Visualization and Computer Graphics (IEEE VIS 2025), Conditional Accept, 2026. }Jun-Hsiang Yao, Xinfang Tian, Jielin Feng, Gulshat Amirkhanova, Kai Xu, Siming Chen*. Beyond the Broadcast: Enhancing VR Tennis Broadcasting through Embedded Visualizations and Camera Techniques. IEEE Transactions on Visualization and Computer Graphics (IEEE VIS 2025), Conditional Accept, 2026. }Yifei Zhang, Lin-Ping Yuan, Yuheng Zhao, Jielin Feng, Siming Chen*. KinemaFX: A Kinematic-Driven Interactive System for Particle Effect Exploration and Customization. In Proceedings of ACM UIST 25, Conditional Accept. FDUVIS 79

80. 上海市数据科学重点实验室复旦大学大数据学院陈思明青年研究员，博导数据可视化将数据转换为图形图像等可视的形式，方便用户快速发掘和理解数据里面的故事，在数据和用户之间架起桥梁，是大数据分析的最后一公里。研究方向一：领域驱动的大模型驱动的智能可视分析 • 构建数据分析智能体作为分析助手，自动进行任务分解与规划，嵌入人机交互的可视化环境进行高效、快捷、低代码数据分析 • 根据用户需求自动生成可视化，组合成可视分析系统，提供满足人类需求的智能分析研究方向二：大模型驱动的智能可视化故事叙述 • 构建大模型驱动的智能可视化设计，快速对数据进行生成与表达，生成满足需求图形 • 构建大模型驱动的数据视频，通过动态的方式生动、灵活展示数据研究方向三：数字孪生与沉浸式可视化 • 支持VR、AR结合可视化与场景渲染，沉浸式数据分析 http://fduvis.net simingchen3@gmail.com FDUVIS 陈思明，复旦大学大数据学院青年研究员，博士生导师，上海市高层次引进人才，复旦学士，北大博士，从事大数据可视化与可视分析的研究，共发表论文 100 余篇，其中在 IEEE VIS ， IEEE TVCG, ACM CHI, CSCW，UIST等顶级国际可视化与人机交互会议以及期刊(CCF A)上发表40余篇文章。曾获评AI2000十年间国际可视化研究最有影响力提名奖 (全球100名)，担任IEEE VIS 国际程序委员会委员， IEEE CG&A国际期刊副主编，IEEE PacificVis论文（VizNotes短文）主席等。应用落地领域自动驾驶金融科技舆情分析人文社会科学大数据 80

81. 复旦大学可视分析与智能决策实验室（FDUVIS） h(p://fduvis.net/ •复旦大学可视分析与智能决策实验室（FDU-VIS）成立于2020年9月，我们的研究方向涵盖了可视化与可视分析、人机混合智能、用户行为分析、决策支持与数据新闻故事叙述等，并在多个应用领域，包括社交媒体、网络空间安全、时空城市大数据、人文历史数据与金融科技方面有一定的科研成果。 • •联系方式： simingchen3@gmail.com