Pith. sign in

REVIEW 3 major objections 4 minor 59 references

VisCanvas: A Node-based Interface for Exploratory Visualization Authoring with LLMs

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read VisCanvas shows that node-based LLM authoring produces more diverse exploration than chat.

desk verdict Solid system, careful study, but the 'more diverse data interaction' headline is overstated: the topology measure is largely a function of the interface's own affordances. read the letter →

arxiv 2607.21886 v1 pith:OEWF6TKL submitted 2026-07-24 cs.HC

Yuki Ueno , Bretho Danzy III , Zhuojun Jiang , Chris Bryan This is my paper
classification cs.HC
keywords visualizationauthoringlargelanguagemodelsexploratoryanalysisnode-basedinterfaceprovenanceVega-Liteuserstudynaturalinterfaces
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that chat-based interfaces are a poor fit for exploratory visualization authoring, where users need to branch, compare, and revisit partial ideas. VisCanvas replaces the linear chat thread with a node graph in which each visualization is a reusable, editable state, and adds operations to modify, duplicate, merge, and suggest. In a controlled 20-person study, the node-based interface produced more Hybrid and Tree exploration structures and fewer Linear ones than a matched chat interface, with workload and usability statistically indistinguishable. Participants preferred VisCanvas for open-ended analysis (14/20) while chat remained competitive for targeted tasks. The design principle distilled is that externalizing exploration structure supports parallel reasoning without extra cognitive cost.

What carries the argument

The node graph itself: each Visualization Node stores a Vega-Lite specification plus the LLM's analysis goal and rationale, and edges record how states derive from one another. Operations (Modify, Duplicate, Merge, Suggest, Fill Blank, Branch) let users fork and recombine states. A staged LLM pipeline—data summarizer, analysis-goal generator, and spec generator with self-reflection—converts prompts into compilable specifications, with the upstream analysis goals reused to ground suggestions.

What would settle it

Re-run the study with two independent coders using a pre-registered codebook, report inter-rater agreement per category, and apply a permutation or chi-square test to the topology distribution. If the difference between VisCanvas and the chat baseline in Hybrid/Tree vs Linear counts does not reach significance (or the agreement on Hybrid falls below a conventional threshold), the diversity claim reduces to anecdote.

Watch

Extended reading notes

Core claim

The central claim is that making visualization states first-class nodes on a graph—rather than messages in a chat—changes how people explore data with LLMs. In the study, VisCanvas users branched and merged more (Hybrid 9 vs 4, Tree 5 vs 2, Linear 1 vs 8), and the difference could not be attributed to better suggestions, since VisChat used the same suggestion backend. The paper interprets this as evidence that node-based interfaces make alternative directions more discoverable and more reusable, turning generated charts into starting points for further analysis.

Load-bearing premise

That the manually coded exploration-topology taxonomy is a valid, reliable measure of 'diverse data interaction'; the Hybrid class—the category driving the diversity claim—had only 62% initial coder agreement, and no significance test is reported for the topology counts.

Editorial extensions

If this is right

  • A node-based interface can support exploratory visual analysis with LLMs without increasing cognitive load relative to chat.
  • Users prefer node-based interaction for open-ended exploration, while chat remains preferred for targeted, hypothesis-driven tasks.
  • Suggestions become more actionable when they persist as nodes rather than ephemeral chat text, since they accounted for 40.7% of branch targets leading to visualizations in VisCanvas.
  • Externalizing intermediate states as reusable specifications enables downstream reuse in other authoring workflows.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The headline diversity result rests on a manually coded topology taxonomy whose key category (Hybrid) had only 62% initial inter-rater agreement and was not tested for significance; a pre-registered re-analysis with an inferential test would settle whether node-based interfaces genuinely change exploration structure.
  • If the effect is real, it likely transfers to other generative-authoring domains (e.g., creative coding, prompt engineering) where users need to fork and merge partial artifacts.
  • A hybrid interface—chat for focused refinement plus a node graph for branching—might capture both preferences, since 10/20 participants preferred chat for targeted tasks.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper presents VisCanvas, a node-based interface for LLM-assisted exploratory visualization authoring, and compares it against VisChat, a chat-based baseline with matched capabilities (Suggest, Visual Builder, Vega-Lite Editor). A within-subjects study with 20 graduate CS students found that participants rated VisCanvas favorably for open-ended tasks and that VisCanvas produced more Hybrid and Tree exploration topologies, with no significant differences in NASA-TLX workload or task success. The paper claims VisCanvas facilitates more diverse data interaction while maintaining comparable performance, and distills design principles for non-linear, provenance-inspired LLM interfaces.

Significance. If the central claim is established, this is a useful contribution to the growing line of work on LLM-based visualization authoring: it provides a concrete node-based interaction design, a feature-matched controlled comparison, and a systematic coding of exploration topologies. The paper ships reproducible artifacts (OSF study materials, GitHub codebase) and uses a counterbalanced within-subjects design with a baseline that controls for several confounds. The preference data and qualitative findings are plausible and informative. However, the headline claim of 'more diverse data interaction' currently relies on descriptive topology proportions without inferential statistics, and the topology measure is tightly coupled to VisCanvas's structural affordances (branch/merge operations) while content-level diversity measures show no difference. The contribution is real but the headline claim needs substantial reframing or additional evidence.

major comments (3)
  1. [§5.2.4] The diversity claim is supported only by descriptive topology counts (Hybrid 9 vs 4, Tree 5 vs 2, Linear 1 vs 8) with no significance test, effect size, or confidence interval. With N=20 and small cell counts, these differences could easily arise by chance. A permutation test or exact test on the contingency table should be reported, along with a measure of uncertainty, before claiming that VisCanvas 'facilitates more diverse data interaction' (Abstract, §7). As written, the headline result is not statistically supported.
  2. [Appx F / §5.2.4] The exploration-topology taxonomy is load-bearing for the diversity claim, but its validity is questionable. First, the Hybrid category—the category driving the between-interface difference—had only 62% initial inter-rater agreement, the lowest of all categories. Second, the taxonomy is derived directly from the graph/chat structure, which is not independent of the interface: VisCanvas explicitly provides Branch, Merge, Duplicate, and Suggest operations that create tree/hybrid structures, while VisChat is a linear chat interface with no mechanism for forking or merging threads. The topology distribution may therefore reflect representational affordances rather than diversity of analytical engagement. This confound should be acknowledged and addressed, e.g., by reporting content-level diversity measures side-by-side and by softening causal wording.
  3. [§5.2.4] The paper reports no significant between-interface differences in Visualization Nodes created or unique variable sets explored. Unique variable sets is a more direct content-level measure of analytical diversity, and it shows no VisCanvas advantage. The manuscript does not attempt to reconcile this null result with the abstract's 'more diverse data interaction' claim. The claim should either be narrowed to 'more structurally divergent exploration patterns' or supported by additional evidence that this structural divergence corresponds to richer user-driven analysis rather than interface mechanics.
minor comments (4)
  1. [§5.2.4 / Fig. 12] Figure 12 is referenced as showing the distribution by interface, but the main text already lists the numbers; adding exact per-cell counts in the figure would improve transparency. Also, the 'Tree Parallel' category has only one observation; its interpretation ('only VisCanvas produced it') should be flagged as anecdotal.
  2. [§5.1] Please define clearly how 'unique variable sets' was computed from logs (e.g., sets of fields used in Vega-Lite encodings across all visualization nodes). This measure is mentioned but not operationalized.
  3. [Appx F] The agreement percentages for each category are informative, but reporting Cohen's kappa or Krippendorff's alpha—rather than only raw agreement—would strengthen the reliability claim.
  4. [Abstract / §7] The wording 'facilitates more diverse data interaction' is stronger than the presented evidence. Consider replacing with a formulation such as 'supports structurally more varied exploration patterns' and explicitly acknowledging the lack of a significance test.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the evaluation is an external empirical comparison, with no fitted parameter being renamed as a prediction or load-bearing self-citation.

full rationale

The paper's central claim—that VisCanvas supports more diverse exploration—is supported by logged user behavior (exploration topologies), self-report (NASA-TLX, Likert), and freeform feedback. The topology taxonomy is manually coded from participant logs rather than derived from any equation or fitted model, and no quantity in the results is defined in terms of the system's own outputs in a way that forces the conclusion. The design considerations (C1–C3) are presented as design goals distilled from prior work, and the evaluation tests them empirically; the fact that the node-based interface implements branching and merging operations and then produces more branched topologies is an empirical observation, not a definitional identity. The paper also includes appropriate controls (VisChat uses the same suggestion backend), and clearly discloses limitations (e.g., small N, graduate-student participants, ceiling effects, Hybrid coding agreement of 62%). No load-bearing argument rests on a self-citation. Accordingly, no circular step meeting the quoted-reduction standard was identified.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

All load-bearing assumptions are methodological rather than mathematical: topology coding validity, equivalence interpretation of null p-values, and generalizability of the participant pool. The paper itself acknowledges the generalizability and equivalence concerns (§6). No numerical free parameters or invented entities appear.

assumptions (4)
  • domain assumption Linear chat interfaces are a poor fit for exploratory visual analysis.
    Assumed in the Introduction and Design Considerations; motivates the system but is not tested directly as a standalone hypothesis.
  • ad hoc to paper The authors' exploration-topology taxonomy validly captures 'diversity of data interaction.'
    Developed by the authors (Appx F); Hybrid category had 62% initial agreement; counts are descriptive only and drive the main claim.
  • domain assumption NASA-TLX and 7-point Likert items measure cognitive load and usability at interval scale.
    Used in §5.2.2 with Wilcoxon signed-rank tests; interpreting null results as 'indistinguishable' assumes adequate power and equivalence.
  • domain assumption Vega-Lite's grammar covers the chart types needed for exploratory analysis.
    The system is limited to Vega-Lite; acknowledged as a limitation in §6.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VisCanvas: A Node-based Interface for Exploratory Visualization Authoring with LLMs." pith.science (2026). https://pith.science/paper/OEWF6TKL

@misc{pith2026260721886,
  author       = {Pith},
  title        = {Pith review of: VisCanvas: A Node-based Interface for Exploratory Visualization Authoring with LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OEWF6TKL}},
  note         = {Machine review of arXiv:2607.21886}
}
read the original abstract

Visual data analysis involves both open-ended exploration and targeted question answering. Visualization authoring tools support this process by enabling users to create visualizations for these tasks. With the rise of large language models, substantial effort has been devoted to developing visualization authoring tools that use natural language instructions. However, existing systems are typically based on a linear chat interface, which is not well suited to exploratory visual analysis workflows. In this paper, we introduce VisCanvas, a node-based interface for exploratory visualization authoring with LLMs. VisCanvas allows users to create, revise, branch, and merge visualizations in a non-linear way, enabling more efficient exploration of multiple analytical directions. We conducted a user study with 20 participants to evaluate the effectiveness of VisCanvas compared to a baseline chat-based interface. The results show that VisCanvas facilitates more diverse data interaction while maintaining performance levels (i.e., cognitive load and usability) that are indistinguishable from current prevailing methods. We then distill design principles for future AI-assisted visualization authoring environments. All supplemental materials required to reproduce the study are available at https://osf.io/gsxhn/overview?view_only=98e94f52985c4cc2ad32209db8772058.

Figures

Figures reproduced from arXiv: 2607.21886 by the authors.

Figure 1
Figure 1. VISCANVAS supports exploratory visualization authoring through a node-based interface comprising: (a) a Context Menu for adding Text Input Nodes; (b) a Text Input Node for generating and updating visualizations, with (b-1) a column-name autocompletion feature and (b-2) a model selector; (c) a Visualization Node for viewing, interpreting, editing, and bookmarking results; (d) a Bookmark Panel for displaying bookmarke… view at source ↗
Figure 3
Figure 3. (a) Modify enables users to generate visualizations via nat￾ural language; (b) Duplicate clones the parent visualization; (c) Merge combines two distinct data views; (d) Suggest recommends 3–5 LLM￾generated prompts to users. visualizations, then generate a Vega-Lite specification that makes the reasoning between them more explicit (see Appx. A.2). The generated node is inserted between the two existing nodes as a ne… view at source ↗
Figure 2
Figure 2. Visualization Node: (a) Robot icon shows the rationale for the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: (a) Fill Blank inserts an intermediate visualization between two connected nodes; (b) Branch creates a Menu Node or Visualization Node based on the edge’s target node. Semantic Zooming [C1]: To support navigation as the exploration graph grows in size, VISCANVAS employ…
Figure 5
Figure 5. Figure 5: Semantic Zoom: Users can navigate large workflows by adjust [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Data Explorer: Users can explore the datasets by (a) data [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: VISCHAT, a controlled chat-based authoring baseline that preserves the familiar linear layout of commercial LLM chat interfaces while including suggestion and visualization-editing capabilities. The right panel shows the chart and chat history; the middle and left pane…
Figure 8
Figure 8. Figure 8: Participants self-reported scores for NASA TLX questions, ease [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: System Evaluation Results. Likert scale responses from the post-study survey. Wilcoxon signed-rank [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Representative examples of user exploration strategies, ordered by increasing structural complexity (left to right). Node colors correspond [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Backend system flow: (a) visualization specification generation; (b) analysis goal suggestion; (c) LLM pipeline for Visualization specification [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Distribution of participant topologies by interface. While [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 8 canonical work pages

  1. [1]

    https://www.tableau

    Business Intelligence and Analytics Software. https://www.tableau. com. 1, 2

  2. [2]

    Colab.https://developers.google.com/colab. 2

  3. [3]

    https://www

    Microsoft Excel | Free Online Spreadsheets Software. https://www. microsoft.com/en-us/microsoft-365/excel. 2

  4. [4]

    https://www

    Power BI - Data Visualization | Microsoft Power Platform. https://www. microsoft.com/en-us/power-platform/products/power-bi. 2

  5. [5]

    RAWGraphs.https://www.rawgraphs.io/. 2

  6. [8]

    Bostock, V

    M. Bostock, V . Ogievetsky, and J. Heer. D³ Data-Driven Documents.IEEE Transactions on Visualization and Computer Graphics, 17(12):2301–2309, Dec. 2011. doi: 10.1109/TVCG.2011.185 1, 2

    doi:10.1109/tvcg.2011.185 2011

  7. [9]

    Brehmer and T

    M. Brehmer and T. Munzner. A Multi-Level Typology of Abstract Vi- sualization Tasks.IEEE Transactions on Visualization and Computer Graphics, 19(12):2376–2385, Dec. 2013. doi: 10.1109/TVCG.2013.124 1

    doi:10.1109/tvcg.2013.124 2013

  8. [10]

    Buxton.Sketching User Experiences: Getting the Design Right and the Right Design

    B. Buxton.Sketching User Experiences: Getting the Design Right and the Right Design. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 2007. doi: 10.1016/B978-0-12-374037-3.X5043-3 9

    doi:10.1016/b978-0-12-374037-3.x5043-3 2007

Show all 59 references
  1. [11]

    S. P. Callahan, J. Freire, E. Santos, C. E. Scheidegger, C. T. Silva, and H. T. V o. VisTrails: visualization meets data management. InProceedings of the 2006 ACM SIGMOD international conference on Management of data, pp. 745–747. ACM, Chicago IL USA, June 2006. doi: 10.1145/1...

    2006 doi

  2. [12]

    Camburn, B

    B. Camburn, B. Dunlap, T. Gurjar, C. Hamon, M. Green, D. Jensen et al. A systematic method for design prototyping.Journal of Mechanical Design, 137(8):081102, Aug. 2015. doi: 10.1115/1.4030331 9

    2015 doi

  3. [14]

    Z. Chen, J. Chen, S. Ö. Arik, M. Sra, T. Pfister, and J. Yoon. CoDA: Agentic Systems for Collaborative Data Visualization, Oct. 2025. doi: 10. 48550/arXiv.2510.03194 2

    2025 doi

  4. [15]

    V . Dibia. LIDA: A Tool for Automatic Generation of Grammar-Agnostic Visualizations and Infographics using Large Language Models. In D. Bol- legala, R. Huang, and A. Ritter, eds.,Proceedings of the 61st Annual Meet- ing of the Association for Computational Linguistics (Volume ...

    2023 doi

  5. [16]

    Doboš, N

    J. Doboš, N. J. Mitra, and A. Steed. 3d timeline: Reverse engineering of a part-based provenance from consecutive 3d models.Computer Graphics Forum, 33(2):135–144, May 2014. doi: 10.1111/cgf.12311 3

    2014 doi

  6. [17]

    S. P. Dow, A. Glassco, J. Kass, M. Schwarz, D. L. Schwartz, and S. R. Klemmer. Parallel prototyping leads to better design results, more diver- gence, and increased self-efficacy.ACM Trans. Comput.-Hum. Interact., 17(4):18:1–18:24, Dec. 2011. doi: 10.1145/1879831.1879836 1, 9

    2011

  7. [18]

    Dunne, N

    C. Dunne, N. Henry Riche, B. Lee, R. Metoyer, and G. Robertson. Graph- Trail: analyzing large multivariate, heterogeneous networks while sup- porting exploration history. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI ’12, pp. 1663–1672. Asso...

    2012

  8. [19]

    Goswami, P

    K. Goswami, P. Mathur, R. Rossi, and F. Dernoncourt. PlotGen: Multi- Agent LLM-based Scientific Data Visualization via Multimodal Retrieval Feedback. InCompanion Proceedings of the ACM on Web Conference 2025, WWW ’25, pp. 1672–1676. Association for Computing Machinery, New Yor...

    2025

  9. [20]

    K. Gu, R. Shang, T. Althoff, C. Wang, and S. M. Drucker. How Do Ana- lysts Understand and Verify AI-Assisted Data Analyses? InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, pp. 1–22. Association for Computing Machinery, New York, NY , US...

    2024

  10. [21]

    Y . Guo, H. Shao, C. Liu, K. Xu, and X. Yuan. PrompTHis: Visualizing the Process and Influence of Prompt Editing During Text-to-Image Creation. IEEE Transactions on Visualization and Computer Graphics, 31(9):4547– 4559, Sept. 2025. doi: 10.1109/TVCG.2024.3408255 3

    2025

  11. [22]

    Y . Han, C. Zhang, X. Chen, X. Yang, Z. Wang, G. Yu et al. ChartLlama: A Multimodal LLM for Chart Understanding and Generation, Nov. 2023. doi: 10.48550/arXiv.2311.16483 1, 2

    2023 doi arXiv

  12. [23]

    J. Heer, J. Mackinlay, C. Stolte, and M. Agrawala. Graphical Histories for Visualization: Supporting Analysis, Communication, and Evaluation. IEEE Transactions on Visualization and Computer Graphics, 14(6):1189– 1196, Nov. 2008. doi: 10.1109/TVCG.2008.137 2, 3

    2008 doi

  13. [24]

    Hong and A

    M.-H. Hong and A. Crisan. Data Has Entered the Chat: How Data Workers Conduct Exploratory Visual Analytic Conversations with GenAI Agents. ACM Trans. Interact. Intell. Syst., 15(4):21:1–21:40, Dec. 2025. doi: 10. 1145/3744750 2

    2025

  14. [25]

    Hutchinson, R

    M. Hutchinson, R. Jianu, A. Slingsby, and P. Madhyastha. Foundation model assisted visual analytics: Opportunities and Challenges.Computers & Graphics, 130:104246, Aug. 2025. doi: 10.1016/j.cag.2025.104246 2

    2025

  15. [26]

    Javed and N

    W. Javed and N. Elmqvist. ExPlates: Spatializing Interactive Analysis to Scaffold Visual Exploration.Computer Graphics Forum, 32(3pt4):441– 450, June 2013. doi: 10.1111/cgf.12131 3

    2013 doi

  16. [27]

    J. Kim, S. Suh, L. B. Chilton, and H. Xia. Metaphorian: Leveraging Large Language Models to Support Extended Metaphor Creation for Science Writing. InProceedings of the 2023 ACM Designing Interactive Systems Conference, DIS ’23, pp. 115–135. Association for Computing Machinery...

    2023

  17. [28]

    Kluyver, B

    T. Kluyver, B. Ragan-Kelley, F. Pérez, B. Granger, M. Bussonnier, J. Fred- eric et al. Jupyter Notebooks—a publishing format for reproducible computational workflows. InIOS Press, pp. 87–90, 2016. doi: 10.3233/ 978-1-61499-649-1-87 2

    2016

  18. [30]

    Z. Liu, J. Thompson, A. Wilson, M. Dontcheva, J. Delorey, S. Grigg et al. Data Illustrator: Augmenting Vector Design Tools with Lazy Data Binding for Expressive Visualization Authoring. InProceedings of the 2018 CHI Conference on Human Factors in Computing Systems, CHI ’18, pp...

    2018

  19. [31]

    S. L’Yi, Q. Wang, F. Lekschas, and N. Gehlenborg. Gosling: A Grammar- based Toolkit for Scalable and Interactive Genomics Data Visualization. IEEE transactions on visualization and computer graphics, 28(1):140–150, Jan. 2022. doi: 10.1109/TVCG.2021.3114876 1, 2

    2022

  20. [32]

    K.-L. Ma. Image graphs-a novel approach to visual data exploration. InProceedings Visualization ’99 (Cat. No.99CB37067), pp. 81–88, Oct

  21. [33]

    Maddigan and T

    P. Maddigan and T. Susnjak. Chat2VIS: Generating Data Visualizations via Natural Language Using ChatGPT, Codex and GPT-3 Large Lan- guage Models.IEEE Access, 11:45181–45193, May 2023. doi: 10.1109/ ACCESS.2023.3274199 1, 2

    2023

  22. [34]

    Moreira, L

    G. Moreira, L. Ferreira, C. Veiga, M. Hosseini, and F. Miranda. Urbanite: A Dataflow-Based Framework for Human-AI Interactive Alignment in Urban Visual Analytics.IEEE Transactions on Visualization and Com- puter Graphics, 32(01):1065–1075, Jan. 2026. doi: 10.1109/TVCG.2025. 3634644 2

    2026 doi

  23. [35]

    G. G. Méndez, M. A. Nacenta, and S. Vandenheste. iV oLVER: Interactive Visual Language for Visualization Extraction and Reconstruction. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, CHI ’16, pp. 4073–4085. Association for Computing Machinery,...

    2016

  24. [36]

    D. Park, M. Suhail, M. Zheng, C. Dunne, E. Ragan, and N. Elmqvist. StoryFacets: A design study on storytelling with visualizations for collab- orative data analysis.Information Visualization, 21(1):3–16, Jan. 2022. doi: 10.1177/14738716211032653 3, 9

    2022 doi

  25. [38]

    D. Ren, B. Lee, and M. Brehmer. Charticulator: Interactive Construction of Bespoke Chart Layouts.IEEE Transactions on Visualization and Computer Graphics, 25(1):789–799, Jan. 2019. doi: 10.1109/TVCG.2018 .2865158 1

    2019 doi

  26. [39]

    Satyanarayan and J

    A. Satyanarayan and J. Heer. Lyra: An Interactive Visualization Design Environment.Computer Graphics Forum, 33(3):351–360, June 2014. doi: 10.1111/cgf.12391 1

    2014 doi

  27. [40]

    Satyanarayan, B

    A. Satyanarayan, B. Lee, D. Ren, J. Heer, J. Stasko, J. Thompson et al. Critical Reflections on Visualization Authoring Systems.IEEE Transac- tions on Visualization and Computer Graphics, 26(1):461–471, Jan. 2020. doi: 10.1109/TVCG.2019.2934281 1, 2

    2020

  28. [41]

    Satyanarayan, D

    A. Satyanarayan, D. Moritz, K. Wongsuphasawat, and J. Heer. Vega-Lite: A Grammar of Interactive Graphics.IEEE Transactions on Visualization and Computer Graphics, 23(1):341–350, Jan. 2017. doi: 10.1109/TVCG. 2016.2599030 1, 2

    2017

  29. [42]

    Satyanarayan, R

    A. Satyanarayan, R. Russell, J. Hoffswell, and J. Heer. Reactive Vega: A Streaming Dataflow Architecture for Declarative Interactive Visualization. IEEE Transactions on Visualization and Computer Graphics, 22(1):659– 668, Jan. 2016. doi: 10.1109/TVCG.2015.2467091 1, 2

    2016

  30. [43]

    L. Shen, H. Li, Y . Wang, X. Xie, and H. Qu. Prompting Generative AI with Interaction-Augmented Instructions. InProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, CHI EA ’25, pp. 1–9. Association for Computing Machinery, New York...

    2025

  31. [44]

    L. Shen, E. Shen, Y . Luo, X. Yang, X. Hu, X. Zhang et al. Towards Natural Language Interfaces for Data Visualization: A Survey.IEEE Transactions on Visualization and Computer Graphics, 29(6):3121–3144, June 2023. doi: 10.1109/TVCG.2022.3148007 2

    2023

  32. [45]

    Shneiderman

    B. Shneiderman. Creativity support tools: accelerating discovery and innovation.Commun. ACM, 50(12):20–32, Dec. 2007. doi: 10.1145/ 1323688.1323689 3

    2007

  33. [46]

    Stolte, D

    C. Stolte, D. Tang, and P. Hanrahan. Polaris: a system for query, analysis, and visualization of multidimensional relational databases.IEEE Trans- actions on Visualization and Computer Graphics, 8(1):52–65, Jan. 2002. doi: 10.1109/2945.981851 2

    2002

  34. [47]

    S. Suh, B. Min, S. Palani, and H. Xia. Sensecape: Enabling Multilevel Exploration and Sensemaking with Large Language Models. InProceed- ings of the 36th Annual ACM Symposium on User Interface Software and Technology, UIST ’23, pp. 1–18. Association for Computing Machinery, Ne...

    2023

  35. [48]

    Y . Tian, W. Cui, D. Deng, X. Yi, Y . Yang, H. Zhang et al. ChartGPT: Leveraging LLMs to Generate Charts From Abstract Natural Language. IEEE Transactions on Visualization and Computer Graphics, 31(3):1731– 1745, Mar. 2025. doi: 10.1109/TVCG.2024.3368621 1, 2, 3

    2025

  36. [50]

    Vaithilingam, E

    P. Vaithilingam, E. L. Glassman, J. P. Inala, and C. Wang. DynaVis: Dynamically Synthesized UI Widgets for Visualization Editing. InPro- ceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI ’24, pp. 1–17. Association for Computing Machinery, New York,...

    2024

  37. [51]

    X. Wang, C. Liang, S. Zheng, J. Liang, G. Li, Y . Zhang et al. Visualization Generation with Large Language Models: An Evaluation, Dec. 2025. doi: 10.48550/arXiv.2401.11255 2, 5

    2025 doi

  38. [52]

    Z. J. Wang, D. Munechika, S. Lee, and D. H. Chau. SuperNOV A: Design Strategies and Opportunities for Interactive Visualization in Computational Notebooks. InExtended Abstracts of the CHI Conference on Human Factors in Computing Systems, CHI EA ’24, pp. 1–17. Association for C...

    2024

  39. [53]

    Z. Wen, L. Weng, Y . Tang, R. Zhang, Y . Liu, B. Pan et al. Exploring Multimodal Prompt for Visualization Authoring with Large Language Models.IEEE Transactions on Visualization and Computer Graphics,

  40. [54]

    R. W. White and R. A. Roth.Exploratory search: Beyond the query- response paradigm. Number 3. Morgan & Claypool Publishers, 2009. doi: book/10.1007/978-3-031-02260-9 3

    2009 doi

  41. [55]

    Wickham.ggplot2: Elegant Graphics for Data Analysis

    H. Wickham.ggplot2: Elegant Graphics for Data Analysis. Springer- Verlag New York, 2016. doi: 10.1007/978-3-319-24277-4 1, 2

    2016 doi

  42. [56]

    Wilkinson.The Grammar of Graphics

    L. Wilkinson.The Grammar of Graphics. Springer Berlin Heidelberg, Berlin, Heidelberg, 2012. doi: 10.1007/978-3-642-21551-3_13 2

    2012 doi

  43. [57]

    Wongsuphasawat, D

    K. Wongsuphasawat, D. Moritz, A. Anand, J. Mackinlay, B. Howe, and J. Heer. V oyager: Exploratory Analysis via Faceted Browsing of Visu- alization Recommendations.IEEE Transactions on Visualization and Computer Graphics, 22(1):649–658, Jan. 2016. doi: 10.1109/TVCG.2015 .2467191 2, 3

    2016 doi

  44. [58]

    Wongsuphasawat, Z

    K. Wongsuphasawat, Z. Qu, D. Moritz, R. Chang, F. Ouk, A. Anand et al. V oyager 2: Augmenting Visual Analysis with Partial View Specifica- tions. InProceedings of the 2017 CHI Conference on Human Factors in Computing Systems, CHI ’17, pp. 2648–2659. Association for Comput- ing...

    2017 doi

  45. [59]

    K. Xu, A. Ottley, C. Walchshofer, M. Streit, R. Chang, and J. Wenskovitch. Survey on the Analysis of User Interactions and Visualization Provenance. Computer Graphics Forum, 39(3):757–783, July 2020. doi: 10.1111/cgf. 14035 3

    2020 doi

  46. [60]

    Yu and C

    B. Yu and C. T. Silva. VisFlow - Web-based Visualization Framework for Tabular Data with a Subset Flow Model.IEEE Transactions on Vi- sualization and Computer Graphics, 23(1):251–260, Jan. 2017. doi: 10. 1109/TVCG.2016.2598497 3

    2017

  47. [61]

    Yu and C

    B. Yu and C. T. Silva. FlowSense: A Natural Language Interface for Visual Data Exploration within a Dataflow System.IEEE Transactions on Visualization and Computer Graphics, 26(1):1–11, Jan. 2020. doi: 10. 1109/TVCG.2019.2934668 3

    2020

  48. [62]

    Z. Zhou, W. Wang, M. Guo, Y . Wang, and D. Gotz. A Design Space for Surfacing Content Recommendations in Visual Analytic Platforms.IEEE Transactions on Visualization and Computer Graphics, 29(1):84–94, Jan

  49. [1999]

    doi: 10.1109/VISUAL.1999.809871 3

    1999

  50. [2023]

    - Existing Vega-Lite Specifications: {json.dumps(state[’input_vega_specs’], indent=2)}

    doi: 10.1109/TVCG.2022.3209445 3 A PROMPTS FORVISUALIZATIONGENERATION In each prompt, certain variables are enclosed within curly brackets and colored blue for easy recognition. A.1 Prompt for Analysis Goal Generator The following variables are used in this prompt: •state[’dat...

    2022

  51. [2026]

    doi: 10.1109/TVCG.2026.3701510 2

    In press. doi: 10.1109/TVCG.2026.3701510 2

    2026

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.