Pith. sign in

REVIEW 3 major objections 7 minor 67 references

GistVis: Automatic Generation of Word-scale Visualizations from Data-rich Documents

T0 review · 3 major / 7 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read GistVis automatically turns data-heavy paragraphs into tiny in-text charts.

desk verdict A solid exploratory systems paper whose headline effectiveness rests on two uncorrected NASA-TLX subscales; the accuracy gain is not significant. read the letter →

arxiv 2502.03784 v1 pith:P7RP2OUM submitted 2025-02-06 cs.HC

classification cs.HC
keywords word-scalevisualizationautomaticlargelanguagemodelsdata-richdocumentsdocument-centricanalysisdatafactinteractivenaturalprocessing
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

GistVis is an automatic pipeline that reads data-rich documents, finds sentences that carry a data insight, classifies the insight type, extracts the underlying numbers, and draws a small word-scale visualization in place. The paper claims this can be done well enough with large language models plus visualization design rules that readers understand documents better and with less mental effort. The main evidence is a user study in which 12 participants answered questions about six news articles, once with plain text and once with GistVis; accuracy was 5.6 points higher, while self-reported mental demand and effort dropped significantly. The deeper claim is that automatic word-scale visualization is feasible as a document-centric augmentation, not just a visualization-centric one.

What carries the argument

The central object is the data fact, an intermediate representation that wraps a text unit segment and, optionally, a data specification with four fields: space, breakdown, feature, and value. This uniform structure lets the pipeline treat plain text and data insights alike, so word-scale visualizations are generated from the same unit segments that carry the narration. The mechanism is prompt chaining: a large language model segments paragraphs, classifies each segment's insight type through a two-stage checker-then-moderator design, and extracts structured data, after which a rule-driven Visualizer maps the fact to one of 14 word-scale chart variants and links visual elements back to entities in the text.

What would settle it

Take a professionally written data-rich document where one data insight is deliberately spread across multiple paragraphs, run GistVis, and check whether it fails to produce any visualization for that insight; a single such failure in a naturalistic article would show the core assumption is not universal, and a corpus count of cross-paragraph insights would quantify how often GistVis would miss them.

Watch

Extended reading notes

Core claim

GistVis decomposes text-to-visualization into four modules: Discoverer (paragraph to unit segments), Annotator (label each segment with one of six data fact types), Extractor (fill a data fact specification with space, breakdown, feature, and value), and Visualizer (map data facts to 14 chart variants). The first three modules are chained large language model prompts; the fourth uses hard-coded visualization design rules. The paper reports that the Discoverer perfectly segments 68.6% of paragraphs, beating regex (54.5%), BERT (61.1%), and Sentence-BERT (60.0%), and the two-stage Annotator reaches 0.79 accuracy and 0.84 F1. In the user study, reading with GistVis significantly reduced mental demand (p=0.016) and perceived effort (p=0.033) compared to plain text, while accuracy trended higher.

Load-bearing premise

The pipeline assumes that every data insight is described by adjacent sentences within a single paragraph, so paragraph-level segmentation can isolate a complete insight without looking across paragraphs.

Editorial extensions

If this is right

  • Data-rich journalism articles can be augmented with inline mini-charts automatically, without hand-authored visualizations.
  • GistVis supports document-centric reading: interaction logs show that readers visit the word-scale visualizations that contain the answer to the task at hand.
  • The two-stage Annotator design (type checker plus type moderator) is a reusable pattern for classifying sparse, imbalanced text segments with LLMs.
  • Because the Extractor and Visualizer stages are deterministic, the pipeline's output is predictable and the uncertainty is concentrated in the LLM-driven stages.
  • The modular four-stage architecture makes it possible to swap in different language models or new chart designs without changing the overall data-fact representation.

Reading between the lines

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

  • The paragraph-level segmentation assumption implies GistVis would miss insights that are spread across paragraphs; a document-level discovery stage that merges related paragraphs would be a natural extension.
  • The accuracy gain of 5.6% was not statistically significant, so the utility argument rests mainly on the workload reductions; a larger replication study could test whether the accuracy trend holds.
  • The same prompt-chaining pattern could generalize to scientific papers or business reports, but the formative corpus covers only data journalism, so the narrative-feature findings may not transfer.
  • The interaction logs suggest readers look at the mini-chart neighboring the answer; eye-tracking could turn this suggestive pattern into direct evidence about how the visualizations guide attention.
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 / 7 minor

Summary. The paper proposes GistVis, an automatic pipeline that transforms data-rich documents into word-scale visualizations through four modules: Discoverer (paragraph-to-unit-segment segmentation), Annotator (fact-type classification), Extractor (data-specification extraction), and Visualizer (chart rendering). A formative corpus study of 44 data journalism articles motivates design constraints, and a technical evaluation reports 68.6% segmentation accuracy and a weighted F1 of 0.84 for the Annotator. A user study (N=12) compares reading with GistVis versus plain text on six Pew Research articles, reporting a non-significant accuracy improvement (+5.6%, p=0.182), no significant time difference, and significant reductions on two NASA-TLX subscales (mental demand p=0.016, effort p=0.033). The abstract claims GistVis is effective in facilitating understanding and reducing workload.

Significance. If fully supported, GistVis would be a useful contribution to automatic visualization generation and document-centric reading. The paper's strengths are its clear modular architecture, the explicit data-fact representation, the use of LLM prompt chaining with documented prompts, and an unusually candid discussion of assumptions and limitations. The code and prompts are made available, which supports reproducibility. However, the central quantitative evidence is fragile: the workload reductions do not survive multiple-comparison correction, the accuracy gain is not significant, and the technical evaluation uses the authors' own annotations as ground truth. The paper is best read as a design study with exploratory evaluation rather than a rigorously validated effectiveness claim.

major comments (3)
  1. [Sec. 6.7.4 / Sec. 6.4] The headline evidence for effectiveness—significant reductions in mental demand (p=0.016) and effort (p=0.033)—is statistically fragile. Section 6.4 states that the NASA-TLX was administered only after the last two articles, once per condition per participant, from different articles. Section 6.7.4 reports unadjusted Wilcoxon p-values across six subscales. Applying a standard multiple-comparison correction (e.g., Holm-Bonferroni with α=0.05 over six tests) sets the threshold for the smallest p at 0.0083, so neither p=0.016 nor p=0.033 remains significant. The paper also does not report which articles were paired with which condition or a per-article analysis, so article difficulty is a plausible confound. The abstract's claim of 'significantly reducing mental demand (p=0.016) and perceived effort (p=0.033)' is therefore not supported by the reported analysis. Please provide corrected p-values, effect sizes, and a mixed-model or per-article analysis, and soften the abstract if the effects do not survive.
  2. [Sec. 5.1.1 / Sec. 5.2.1 / Sec. 3.2] The technical evaluation is circular in a load-bearing way. The ground-truth unit segments and fact-type annotations were produced by the same researchers who designed the GistVis pipeline, using a fact taxonomy that the authors themselves pruned and operationalized. Sections 5.1.1 and 5.2.1 report accuracy and F1 scores against this self-created corpus, with no inter-rater reliability, no independent second annotation, and no external validation. This measures agreement with the designers' interpretation rather than with an objective standard. Moreover, the Extractor and Visualizer modules are not directly evaluated at all (Section 5.1 states they are excluded), and the user study does not isolate their errors from those of the Discoverer and Annotator. To support the claim of 'decent performance', the authors should report annotation reliability, test on an independently annotated corpus (or have external judges label a held-out set), and provide at least a sanity-check evaluation of extracted data specifications and generated visualizations.
  3. [Sec. 4.1.1 / Sec. 7] The paragraph-proximity assumption substantially narrows the system's scope relative to the paper's framing. Section 4.1.1 states the assumption that all information for one data insight is sequential, proximate, and within the same paragraph, and Section 7 concedes that insights can span paragraphs or the entire passage. The user study uses six articles that apparently fit this assumption, but the title and abstract claim automatic generation from 'data-rich documents' in general. Since cross-paragraph insights are missed entirely by the pipeline, the paper should either explicitly restrict its contribution to this class of documents or provide corpus evidence about how frequently the assumption holds. As written, the generality of the central claim is not established.
minor comments (7)
  1. [Sec. 6.7.2] The +5.6% accuracy difference is described as a non-significant trend in Section 6.7.2 but is presented without qualification in the abstract. Please consistently label it as not statistically significant.
  2. [Fig. 7 / Fig. 9] The scatter plots with divergent color scales obscure the paired structure of the data; paired strip charts or boxplots with per-participant lines would make the comparisons easier to interpret.
  3. [Sec. 4.2.4] The rank-visualization limit of 10 is a free parameter justified by a corpus observation. Since it may not transfer to other corpora, it should be explicitly reported as a tunable parameter rather than an inherent design constraint.
  4. [Sec. 5.2.2] The conclusion that the two-step Annotator is superior to the one-step version rests on a single class (value precision dropping to 0.11 in the one-step condition); no statistical test is reported for the difference in per-class performance. Please add quantitative comparisons across all classes.
  5. [Sec. 5.1.1] The BERT and Sentence-BERT thresholds were selected via grid search on the same evaluation corpus used for the reported results. This should be stated explicitly as in-sample tuning, which in this case makes the comparison conservative but needs to be transparent.
  6. [Sec. 3.1 / Sec. 6.3] The corpus consists only of data journalism, and the user study uses only Pew Research articles. The paper acknowledges this, but the generality of the findings to business reports and academic papers remains untested and should be stated as a limitation in the conclusion.
  7. [Throughout] There are several typos and awkward phrasings, e.g., 'Content is where we store' in Section 4.1.1 and 'Over or under-segmentation of sentences without data insights would not impact the visualization result' in Section 5.1.1. A careful language pass is recommended.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the technical evaluation uses the authors' own annotations as ground truth, but the LLM modules are not fitted to those labels, and the main effectiveness claim rests on an external user study rather than a self-citation chain.

full rationale

GistVis is a systems paper, so the claimed derivation chain is empirical rather than formal. The central effectiveness claim is supported by a technical evaluation plus a user study with 12 external participants. The technical evaluation (Sec. 5) does use the authors' own annotated corpus from Sec. 3 as ground truth: 'We use the annotated corpus described in Sec. 3 to evaluate the performance of the Discoverer' and 'we assume the segmentation is correct and directly use the labeled unit segments as input.' This is a self-referential benchmark construction, but it is not a circular reduction: the annotations encode the authors' definitions of unit segments and fact types, while the Discoverer and Annotator are zero-shot and few-shot LLM modules that are evaluated against those labels rather than fitted to them. No parameter is estimated from the evaluation data and then renamed as a prediction; the scores measure agreement with the authors' interpretation, not a forced fit. The fact taxonomy itself comes from external prior work [13], and the main effectiveness claim is grounded in a user study with external participants under GistVis versus Plain Text conditions (Sec. 6). The only self-citations ([12], [66]) appear in related-work and future-work contexts and are not load-bearing. The acknowledged paragraph-proximity assumption (Sec. 4.1.1) and the uncorrected NASA-TLX subscale p-values (Sec. 6.7.4) are validity and statistical concerns, not circularity, and per the review rules they belong under correctness risk rather than this pass. No equation or construction reduces the claimed output to the input, so no significant circularity is present.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the validity of the fact taxonomy, the reliability of LLM outputs for natural-language understanding, and the assumption that insights are local to a paragraph. The only hand-tuned parameter in the visualization stage is the rank limit of 10. No new physical or conceptual entities are introduced.

free parameters (1)
  • Rank visualization limit = 10
    Hand-set maximum rank for the rank visualization to avoid overflow; based on the corpus observation that only 1 of 35 rank insights exceeded 10 (Section 4.2.4). This is a design choice, not a fitted model parameter.
assumptions (4)
  • domain assumption The fact taxonomy of Chen et al. (2009) is a valid representation of data insights in text.
    The authors use this taxonomy to define fact types and annotation criteria without validating its completeness for natural-language data documents.
  • domain assumption Large language models can perform zero-shot segmentation, classification, and data extraction reliably enough for the pipeline.
    The pipeline depends on DeepSeek-V2.5 outputs for all but the final visualization step; errors propagate downstream.
  • domain assumption Data insights are contained within a single paragraph and are described in sequential, proximate sentences.
    Explicitly stated in Section 4.1.1 as a design constraint; if false, many real documents would not be handled.
  • domain assumption Word-scale visualizations are effective for document-centric reading.
    Borrowed from prior literature (Goffin et al.), not independently re-tested except via the small user study.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GistVis: Automatic Generation of Word-scale Visualizations from Data-rich Documents." pith.science (2026). https://pith.science/paper/P7RP2OUM

@misc{pith2026250203784,
  author       = {Pith},
  title        = {Pith review of: GistVis: Automatic Generation of Word-scale Visualizations from Data-rich Documents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P7RP2OUM}},
  note         = {Machine review of arXiv:2502.03784}
}
read the original abstract

Data-rich documents are ubiquitous in various applications, yet they often rely solely on textual descriptions to convey data insights. Prior research primarily focused on providing visualization-centric augmentation to data-rich documents. However, few have explored using automatically generated word-scale visualizations to enhance the document-centric reading process. As an exploratory step, we propose GistVis, an automatic pipeline that extracts and visualizes data insight from text descriptions. GistVis decomposes the generation process into four modules: Discoverer, Annotator, Extractor, and Visualizer, with the first three modules utilizing the capabilities of large language models and the fourth using visualization design knowledge. Technical evaluation including a comparative study on Discoverer and an ablation study on Annotator reveals decent performance of GistVis. Meanwhile, the user study (N=12) showed that GistVis could generate satisfactory word-scale visualizations, indicating its effectiveness in facilitating users' understanding of data-rich documents (+5.6% accuracy) while significantly reducing their mental demand (p=0.016) and perceived effort (p=0.033).

Figures

Figures reproduced from arXiv: 2502.03784 by the authors.

Figure 1
Figure 1. GistVis enables automatic generation of word-scale visualizations from data-rich documents. GistVis supports [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Each element in data specification consists a four [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The GistVis pipeline consists of four modules: Discoverer (M1), Annotator (M2), Extractor (M3), and Visualizer (M4). [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: A collection of 14 candidate visualizations and the corresponding chart type for each data fact type. The Example [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Normalized confusion matrices for data fact type annotation results. The left matrix (A) shows the result of our [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: The interface we employed for our user study. The data-rich document is rendered in the Document Panel on the left. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: A comparison of user performance under GistVis and Plain Text over two utility metrics. Each scatter point in the [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: The aggregated interaction log from all 12 participants over six test passages. The x-axis represents the word-scale [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: A comparison of perceived workload between GistVis and Plain Text over NASA-TLX metrics. Each scatter point in [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 26 canonical work pages

  1. [1]

    R. Amar, J. Eagan, and J. Stasko. 2005. Low-Level Components of Analytic Activity in Information Visualization. In IEEE Symposium on Information Visualization,

  2. [2]

    Narrative Text Vis (NTV) Introduction - AVA

    AVA | AntV 2024. Narrative Text Vis (NTV) Introduction - AVA. Available: https://ava.antv.antgroup.com/en/guide/ntv/intro. Accessed: 2024-08-18

  3. [3]

    Sriram Karthik Badam, Zhicheng Liu, and Niklas Elmqvist. 2019. Elastic Docu- ments: Coupling Text and Tables through Contextual Visualizations for Enhanced Document Reading. IEEE Transactions on Visualization and Computer Graphics 25, 1 (Jan. 2019), 661–671. doi:10.1109/TVCG.2018.2865119

  4. [4]

    Fabian Beck, Sebastian Koch, and Daniel Weiskopf. 2016. Visual Analysis and Dissemination of Scientific Literature Collections with SurVis. IEEE Transactions on Visualization and Computer Graphics 22, 1 (Jan. 2016), 180–189. doi:10.1109/ TVCG.2015.2467757

  5. [5]

    Fabian Beck and Daniel Weiskopf. 2017. Word-Sized Graphics for Scientific Texts. IEEE Transactions on Visualization and Computer Graphics 23, 6 (June 2017), 1576–1587. doi:10.1109/TVCG.2017.2674958

  6. [6]

    O’Reilly Media, Inc

    Steven Bird, Ewan Klein, and Edward Loper. 2009. Natural Language Processing with Python: Analyzing Text with the Natural Language Toolkit . "O’Reilly Media, Inc. "

  7. [7]

    Michael Bostock, Vadim Ogievetsky, and Jeffrey Heer. 2011. D 3 Data-Driven Documents. IEEE Transactions on Visualization and Computer Graphics 17, 12 (Dec. 2011), 2301–2309. doi:10.1109/TVCG.2011.185

  8. [8]

    Ulrik Brandes, Bobo Nick, Brigitte Rockstroh, and Astrid Steffen. 2013. Gestalt- lines. Computer Graphics Forum 32, 3pt2 (2013), 171–180. doi:10.1111/cgf.12104

Show all 67 references
  1. [9]

    Dennis Bromley and Vidya Setlur. 2024. DASH: A Bimodal Data Exploration Tool for Interactive Text and Visualizations. doi:10.48550/arXiv.2408.01011 arXiv:2408.01011 [cs]

  2. [10]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffr...

  3. [12]

    Qing Chen, Ying Chen, Ruishi Zou, Wei Shuai, Yi Guo, Jiazhe Wang, and Nan Cao. 2024. Chart2Vec: A Universal Embedding of Context-Aware Visualizations. IEEE Transactions on Visualization and Computer Graphics (2024), 1–12. doi:10. 1109/TVCG.2024.3383089

  4. [13]

    Yang Chen, Jing Yang, and William Ribarsky. 2009. Toward Effective Insight Management in Visual Analytics Systems. In 2009 IEEE Pacific Visualization Sym- posium. IEEE Computer Society, 49–56. doi:10.1109/PACIFICVIS.2009.4906837

  5. [14]

    Zhutian Chen and Haijun Xia. 2022. CrossData: Leveraging Text-Data Connec- tions for Authoring Data Documents. InCHI Conference on Human Factors in Com- puting Systems. ACM, New Orleans LA USA, 1–15. doi:10.1145/3491102.3517485

  6. [15]

    Patricia Costigan-Eaves. 1986. Edward R. Tufte The Visual Display of Quantitative Information. Information Design Journal 4, 3 (Jan. 1986), 235–236. doi:10.1075/ idj.4.3.12cos

  7. [16]

    Kiana Cox. 2024. black-voters-support-harris-over-trump-and-kennedy-by-a- wide-margin. Available: https://www.pewresearch.org/short-reads/2024/08/22/ black-voters-support-harris-over-trump-and-kennedy-by-a-wide-margin/. Ac- cessed: 2024-09-11. GistVis: Automatic Generation of ...

  8. [17]

    Weiwei Cui, Xiaoyu Zhang, Yun Wang, He Huang, Bei Chen, Lei Fang, Haidong Zhang, Jian-Guan Lou, and Dongmei Zhang. 2020. Text-to-Viz: Automatic Genera- tion of Infographics from Proportion-Related Natural Language Statements. IEEE Transactions on Visualization and Computer Gra...

  9. [18]

    Zhang, Hanwei Xu, Hao Yang, Haowei Zhang, Honghui Ding, Huajian Xin, Huazuo Gao, Hui Li, Hui Qu, J

    DeepSeek-AI, Aixin Liu, Bei Feng, Bin Wang, Bingxuan Wang, Bo Liu, Chenggang Zhao, Chengqi Dengr, Chong Ruan, Damai Dai, Daya Guo, Dejian Yang, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei Li, H. Zhang, Hanwei Xu, Hao Yang, Haowei...

  10. [19]

    Drew DeSilver. 2024. online-shopping-has-grown-rapidly-in-u-s-but-most- sales-are-still-in-stores. Available: https://www.pewresearch.org/short- reads/2023/11/22/online-shopping-has-grown-rapidly-in-u-s-but-most-sales- are-still-in-stores/. Accessed: 2024-09-11

  11. [20]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human...

  12. [21]

    Victor Dibia. 2023. LIDA: A Tool for Automatic Generation of Grammar-Agnostic Visualizations and Infographics Using Large Language Models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), Danushka Bol...

  13. [22]

    Apache ECharts

    echarts 2024. Apache ECharts. Available: https://echarts.apache.org/en/index. html/. Accessed: 2024-04-28

  14. [23]

    Richard Fry. 2024. a-growing-share-of-us-husbands-and-wives-are-roughly-the- same-age. Available: https://www.pewresearch.org/short-reads/2024/08/15/a- growing-share-of-us-husbands-and-wives-are-roughly-the-same-age/. Ac- cessed: 2024-09-11

  15. [24]

    G2Plot: a charting library - AntV

    G2 2024. G2Plot: a charting library - AntV. Available: https://g2plot.antv.antgroup. com/en/. Accessed: 2024-08-18

  16. [26]

    Pascal Goffin, Tanja Blascheck, Petra Isenberg, and Wesley Willett. 2020. Interac- tion Techniques for Visual Exploration Using Embedded Word-Scale Visualiza- tions. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI ’20). Association for Comp...

  17. [27]

    Pascal Goffin, Jeremy Boy, Wesley Willett, and Petra Isenberg. 2017. An Ex- ploratory Study of Word-Scale Graphics in Data-Rich Text Documents. IEEE Transactions on Visualization and Computer Graphics 23, 10 (Oct. 2017), 2275–2287. doi:10.1109/TVCG.2016.2618797

  18. [28]

    Pascal Goffin, Wesley Willett, Anastasia Bezerianos, and Petra Isenberg. 2015. Exploring the Effect of Word-Scale Visualizations on Reading Behavior. InPro- ceedings of the 33rd Annual ACM Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA ’15) . Assoc...

  19. [29]

    Pascal Goffin, Wesley Willett, Jean-Daniel Fekete, and Petra Isenberg. 2014. Ex- ploring the Placement and Design of Word-Scale Visualizations. IEEE Trans- actions on Visualization and Computer Graphics 20, 12 (Dec. 2014), 2291–2300. doi:10.1109/TVCG.2014.2346435

  20. [30]

    Hart and Lowell E

    Sandra G. Hart and Lowell E. Staveland. 1988. Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research. In Advances in Psychology, Peter A. Hancock and Najmedin Meshkati (Eds.). Human Mental Workload, Vol. 52. North-Holland, 139–183. doi:10.1016...

  21. [31]

    Jeffrey Heer and Ben Shneiderman. 2012. Interactive Dynamics for Visual Analy- sis: A Taxonomy of Tools That Support the Fluent and Flexible Use of Visualiza- tions. Queue 10, 2 (Feb. 2012), 30–55. doi:10.1145/2133416.2146416

  22. [32]

    Jane Hoffswell, Arvind Satyanarayan, and Jeffrey Heer. 2018. Augmenting Code with In Situ Visualizations to Aid Program Understanding. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems . ACM, Montreal QC Canada, 1–12. doi:10.1145/3173574.3174106

  23. [33]

    Jessica Hullman, Nicholas Diakopoulos, and Eytan Adar. 2013. Contextifier: Automatic Generation of Annotated Stock Visualizations. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’13). Association for Computing Machinery, New York, NY, USA, 2...

  24. [34]

    Franziska Huth, Maurice Koch, Miriam Awad-Mohammed, Daniel Weiskopf, and Kuno Kurzhals. 2024. Eye Tracking on Text Reading with Visual Enhancements. In Proceedings of the 2024 Symposium on Eye Tracking Research and Applications (Glasgow, United Kingdom) (ETRA ’24). Association...

  25. [35]

    Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran- Johnson, Scott Johnston, Sheer El-Showk, Andy Jones, Nelson Elhage, Tristan Hume, Anna Chen, Yuntao Bai, Sam Bowman, Stanislav Fo...

  26. [36]

    Brian Kennedy. 2024. Most Americans think U.S. K-12 STEM education isn’t above average, but test results paint a mixed picture. Available: https://www.pewresearch.org/short-reads/2024/04/24/most-americans-think- us-k-12-stem-education-isnt-above-average-but-test-results-paint-...

  27. [37]

    Hearst, and Maneesh Agrawala

    Nicholas Kong, Marti A. Hearst, and Maneesh Agrawala. 2014. Extracting Refer- ences between Text and Charts via Crowdsourcing. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’14) . Association for Computing Machinery, New York, NY, USA, 31–4...

  28. [38]

    Sandee LaMotte. 2023. These 8 habits could add up to 24 years to your life, study says. Available: https://edition.cnn.com/2023/07/24/health/habits-live-longer- wellness/index.html. Accessed: 2024-09-12

  29. [39]

    LangChain

    LangChain 2024. LangChain. https://js.langchain.com/v0.1/docs/get_started/ introduction/. Accessed: 2024-08-15

  30. [40]

    Shahid Latif, Zheng Zhou, Yoon Kim, Fabian Beck, and Nam Wook Kim. 2022. Kori: Interactive Synthesis of Text and Charts in Data Documents. IEEE Transactions on Visualization and Computer Graphics 28, 1 (Jan. 2022), 184–194. doi:10.1109/ TVCG.2021.3114802

  31. [41]

    Sukwon Lee, Sung-Hee Kim, and Bum Chul Kwon. 2017. VLAT: Development of a Visualization Literacy Assessment Test. IEEE Transactions on Visualization and Computer Graphics 23, 1 (Jan. 2017), 551–560. doi:10.1109/TVCG.2016.2598920

  32. [42]

    Tianle Li, Ge Zhang, Quy Duc Do, Xiang Yue, and Wenhu Chen. 2024. Long- Context LLMs Struggle with Long In-context Learning. doi:10.48550/arXiv.2404. 02060 arXiv:2404.02060 [cs]

  33. [43]

    Allen Yilun Lin, Joshua Ford, Eytan Adar, and Brent Hecht. 2018. VizByWiki: Mining Data Visualizations from the Web to Enrich News Articles. InProceedings of the 2018 World Wide Web Conference on World Wide Web - WWW ’18 . ACM Press, Lyon, France, 873–882. doi:10.1145/3178876.3186135

  34. [44]

    Alan Lundgard and Arvind Satyanarayan. 2022. Accessible Visualization via Natural Language Descriptions: A Four-Level Model of Semantic Content. IEEE Transactions on Visualization and Computer Graphics 28, 1 (Jan. 2022), 1073–1083. doi:10.1109/TVCG.2021.3114770

  35. [45]

    Damien Masson, Sylvain Malacria, Géry Casiez, and Daniel Vogel. 2023. Chara- graph: Interactive Generation of Charts for Realtime Annotation of Data-Rich Paragraphs. In Proceedings of the 2023 CHI Conference on Human Factors in Com- puting Systems (CHI ’23) . Association for C...

  36. [46]

    Ronald Metoyer, Qiyu Zhi, Bart Janczuk, and Walter Scheirer. 2018. Coupling Story to Visualization: Using Textual Analysis as a Bridge Between Data and Interpretation. In 23rd International Conference on Intelligent User Interfaces. ACM, Tokyo Japan, 503–507. doi:10.1145/31729...

  37. [47]

    Tamara Munzner. 2014. Visualization analysis and design . CRC press

  38. [48]

    Miguel Nacenta, Uta Hinrichs, and Sheelagh Carpendale. 2012. FatFonts: Com- bining the Symbolic and Visual Aspects of Numbers. In Proceedings of the Inter- national Working Conference on Advanced Visual Interfaces (A VI ’12). Association for Computing Machinery, New York, NY, ...

  39. [49]

    Yasmina Okan, Eva Janssen, Mirta Galesic, and Erika A. Waters. 2019. Using the Short Graph Literacy Scale to Predict Precursors of Health Behavior Change.Med- ical decision making : an international journal of the Society for Medical Decision Making 39, 3 (April 2019), 183–195...

  40. [50]

    Saugat Pandey and Alvitta Ottley. 2023. Mini-VLAT: A Short and Effective Measure of Visualization Literacy. Computer Graphics Forum 42, 3 (2023), 1–11. doi:10.1111/cgf.14809

  41. [51]

    Charles Perin, Romain Vuillemot, and Jean-Daniel Fekete. 2013. SoccerStories: A Kick-off for Visual Soccer Analysis. IEEE Transactions on Visualization and Computer Graphics 19, 12 (Dec. 2013), 2506–2515. doi:10.1109/TVCG.2013.192

  42. [52]

    Peity · progressive <svg> pie, donut, bar and line charts

    Piety 2024. Peity · progressive <svg> pie, donut, bar and line charts. https: //benpickles.github.io/peity/. Accessed: 2024-08-16

  43. [53]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Confer- ence on Natural Language Processing (EMNLP-IJ...

  44. [54]

    Katherine Schaeffer. 2024. us-centenarian-population-is-projected-to-quadruple- over-the-next-30-years. Available: https://www.pewresearch.org/short- reads/2024/01/09/us-centenarian-population-is-projected-to-quadruple-over- the-next-30-years/. Accessed: 2024-09-11

  45. [55]

    Sono Shah and Regina Widjaya. 2024. 71-of-asian-restaurants-in-the-u-s-serve- chinese-japanese-or-thai-food. Available: https://www.pewresearch.org/short- reads/2023/05/23/71-of-asian-restaurants-in-the-u-s-serve-chinese-japanese- or-thai-food/. Accessed: 2024-09-11

  46. [56]

    Danqing Shi, Xinyue Xu, Fuling Sun, Yang Shi, and Nan Cao. 2021. Calliope: Automatic Visual Data Story Generation from a Spreadsheet. IEEE Transactions on Visualization and Computer Graphics 27, 2 (Feb. 2021), 453–463. doi:10.1109/ TVCG.2020.3030403

  47. [57]

    Sparklificator by INRIA

    Sparklificator 2024. Sparklificator by INRIA. https://inria.github.io/sparklificator/. Accessed: 2024-08-16

  48. [58]

    Florian Stalph. 2018. Classifying Data Journalism: A Content Analysis of Daily Data-Driven Stories. Journalism Practice 12, 10 (Nov. 2018), 1332–1350. doi:10. 1080/17512786.2017.1386583

  49. [59]

    R Core Team. 2013. R: A language and environment for statistical computing

  50. [60]

    Edward Rolf Tufte. 2006. Beautiful evidence. Vol. 1. Graphics Press Cheshire, CT

  51. [61]

    Yun Wang, Zhida Sun, Haidong Zhang, Weiwei Cui, Ke Xu, Xiaojuan Ma, and Dongmei Zhang. 2020. DataShot: Automatic Generation of Fact Sheets from Tabular Data. IEEE Transactions on Visualization and Computer Graphics 26, 1 (Jan. 2020), 895–905. doi:10.1109/TVCG.2019.2934398

  52. [62]

    J. N. Washburne. 1927. An Experimental Study of Various Graphic, Tabular, and Textual Methods of Presenting Quantitative Material. Journal of Educational Psychology 18, 7 (1927), 465–476. doi:10.1037/h0070054

  53. [63]

    Leland Wilkinson. 2012. The Grammar of Graphics. InHandbook of Computational Statistics: Concepts and Methods , James E. Gentle, Wolfgang Karl Härdle, and Yuichi Mori (Eds.). Springer, Berlin, Heidelberg, 375–414. doi:10.1007/978-3-642- 21551-3_13

  54. [64]

    Aoyu Wu, Yun Wang, Xinhuan Shu, Dominik Moritz, Weiwei Cui, Haidong Zhang, Dongmei Zhang, and Huamin Qu. 2022. AI4VIS: Survey on Artificial Intelligence Approaches for Data Visualization.IEEE Transactions on Visualization and Computer Graphics 28, 12 (Dec. 2022), 5049–5070. do...

  55. [65]

    Tongshuang Wu, Michael Terry, and Carrie Jun Cai. 2022. AI Chains: Transparent and Controllable Human-AI Interaction by Chaining Large Language Model Prompts. InProceedings of the 2022 CHI Conference on Human Factors in Computing Systems (CHI ’22). Association for Computing Ma...

  56. [66]

    Bingsheng Yao, Guiming Chen, Ruishi Zou, Yuxuan Lu, Jiachen Li, Shao Zhang, Yisi Sang, Sijia Liu, James Hendler, and Dakuo Wang. 2024. More Samples or More Prompts? Exploring Effective Few-Shot In-Context Learning for LLMs with In-Context Sampling. In Findings of the Associati...

  57. [67]

    Lau, Jose Echevarria, and Zoya Bylinskii

    Nanxuan Zhao, Nam Wook Kim, Laura Mariah Herman, Hanspeter Pfister, Ryn- son W.H. Lau, Jose Echevarria, and Zoya Bylinskii. 2020. ICONATE: Automatic Compound Icon Generation and Ideation. In Proceedings of the 2020 CHI Con- ference on Human Factors in Computing Systems . ACM, ...

  58. [2005]

    IEEE Computer Society, 111–117

    INFOVIS 2005. IEEE Computer Society, 111–117. doi:10.1109/INFVIS.2005. 1532136

  59. [2022]

    doi:10.48550/arXiv

    Language Models (Mostly) Know What They Know. doi:10.48550/arXiv. 2207.05221 arXiv:2207.05221 [cs]

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.