Pith. sign in

REVIEW 4 major objections 4 minor 78 references

Story Ribbons: Reimagining Storyline Visualizations with Large Language Models

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read An almost fully automated LLM pipeline can extract story data from raw novels and scripts well enough to power interactive storyline visualizations that match human study guides on the vast majority of key characters and events.

desk verdict Solid systems contribution; the headline accuracy numbers need stronger evidence, but the artifact and user study carry it. read the letter →

arxiv 2508.06772 v1 pith:E74WJNXC submitted 2025-08-09 cs.HC cs.CLcs.LG

classification cs.HCcs.CLcs.LG
keywords storylinevisualizationnarrativeliteraryanalysislargelanguagemodelsLLMdataextractioninteractivecharacterandthemetrajectoriestrustcalibration
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 large language models can do the hard data-extraction work that has historically made storyline visualizations of novels and scripts impractical: converting raw text into structured scenes, characters, locations, themes, sentiments, and quotes with enough reliability to drive a usable interactive tool. It introduces Story Ribbons, a visualization system built on an LLM pipeline with correction loops, and reports that on 36 texts the pipeline recovers roughly 90 percent or more of the characters and key events named in human-written study guides. If this holds, literary analysts could trace character and theme trajectories across long books without manual tagging, and LLM-based visualizations could become a standard exploratory layer for reading and teaching literature.

What carries the argument

The load-bearing mechanism is the LLM chaining pipeline, organized as decomposition (splitting the story into chapters, then into scenes defined by location change, with per-scene summaries, ratings, character and theme extraction, and quoted evidence) and aggregation (building chapter, character, and location summaries from the scene data). Two correction loops are essential: an exact-string-match check that replaces hallucinated or modified quotes with LLM explanations, and a second LLM that groups duplicate character and location names. The visualization itself adapts the ribbon storyline technique, with each character as a weighted path whose thickness encodes importance, a y-axis that c

What would settle it

Re-run the study-guide comparison on the same six works with two independent annotators who do the diff-style matching blind to the paper's results; if their agreement (e.g., Cohen's kappa on character and event matches) is below 0.6, or if a gold-standard event list shows that the pipeline's 90% event overlap does not survive an exact-match criterion, the central reliability claim is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that an almost fully automated pipeline—LLM chaining with decomposition into chapters and scenes, aggregation into summaries, and correction loops for hallucinated quotes and duplicated names—can extract narrative data from unstructured novels, plays, and poems well enough to be genuinely useful in practice. On six well-known works, the pipeline matched 94.3% of the characters listed by one popular study guide and 83.3% of those on another, identified 90.4% and 90.1% of the same key events with 100% chronological accuracy, and was the only source to detect roughly 43% of characters and 91% of themes. A 16-participant user study found the resulting Story Ribbons t

Load-bearing premise

The reported reliability numbers rest on the authors' own qualitative matching of LLM outputs to human-written study guides, with no inter-rater reliability check or public annotation set; if that matching is looser than it looks, the claim that the pipeline is 'sufficiently reliable to be helpful' loses its quantitative footing.

Editorial extensions

If this is right

  • A near-automated pipeline can produce structured story data for any novel-length text in minutes (2.2 minutes for The Great Gatsby, 3.9 for Little Women), removing the manual tagging bottleneck that limited prior storyline systems.
  • Because the pipeline is genre- and target-agnostic, the same decomposition and aggregation structure can be pointed at themes, locations, or user-defined traits, enabling custom visualization dimensions without additional data preparation.
  • The correction loops are portable design patterns: exact quote verification and duplicate-entity grouping are generally useful guards for any LLM-based text-extraction task.
  • When LLM outputs are transparent—with explanations and links to the raw text—users can treat disagreements as interpretive prompts rather than errors, as when readers noticed the model rated Mr. Bennet as initially most important in Pride and Prejudice.
  • The system is best used by readers who already know the story; users explicitly said it would spoil discovery for unfamiliar texts, which constrains the tool's intended use cases.

Reading between the lines

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

  • The unblinded, qualitative diff against human study guides is a lenient yardstick; a stricter test would be inter-annotator agreement on a gold-standard corpus, and the exact overlap figures may shift under such a test.
  • The scene-definition choice (location change) is itself an interpretive commitment; the pipeline's 72.7% adherence to that definition shows LLMs impose their own genre-dependent scene semantics, which could be surfaced as a user-controllable parameter.
  • The finding that LLM-only themes vastly outnumber study-guide themes (91.3% LLM-only) suggests a complementary relationship: guides curate, LLMs enumerate; a tool that merges both could offer a criticism layer beyond the text.
  • Because the pipeline relies on two specific commercial LLMs (one for extraction, another for corrections), the reported reliability is tied to those models and may not transfer to open-weight alternatives.
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

4 major / 4 minor

Summary. The paper introduces Story Ribbons, an interactive literary-analysis visualization system, together with an LLM-based pipeline that extracts structured narrative data (chapters, scenes, characters, locations, themes, quotes, and ratings) from novels, plays, poems, and nonfiction. The pipeline decomposes stories into chapters and scenes, applies correction loops for hallucinated quotes and duplicate entities, then aggregates the results into a JSON representation. Story Ribbons visualizes characters and themes as ribbons over time, with customizable y-axes, color encodings, explanations on demand, and natural-language query/dimension features. The authors evaluate the pipeline on 36 texts, compare extracted characters/themes/events against SparkNotes and LitCharts for six canonical works, report quote-accuracy rates, and conduct a 16-participant user study plus expert interviews. The central claim is that, despite known limitations, LLMs can meaningfully augment traditional text visualizations and that the pipeline is 'sufficiently reliable to be helpful to users.'

Significance. If the reliability claims hold, this paper makes a useful contribution to narrative visualization and LLM-supported literary analysis. Its strengths include a rare focus on the full data-extraction pipeline rather than only on layout; the explicit use of correction loops to combat hallucinated quotes and duplicate entities; the inclusion of synthetic novels as a partial control against training-data memorization; evaluation against external human-authored study guides rather than only self-generated ground truth; and transparent reporting of LLM failures in Sec. 7.3. The design process is strengthened by co-design with literary scholars, and the user study provides concrete evidence that the interactive features (explanations, custom dimensions, 'ask LLM') can support interpretation and discussion. The main quantitative support for the 'sufficiently reliable' claim, however, rests on the study-guide overlap analysis in Sec. 4.2.2, which is the paper's weakest point and needs substantial strengthening before the claim is fully established.

major comments (4)
  1. [Sec. 4.2.2, Study Guide Analysis] The 94.3%/83.3% character overlap and 90.4%/90.1% event overlap figures are the primary quantitative basis for the abstract and conclusion claims that the pipeline is 'sufficiently reliable to be helpful.' The method is described only as a 'diff-style comparison' in which characters and themes are 'qualitatively matched when different names likely referred to the same entity.' No matching rubric, no inter-rater reliability, and no released annotation set are provided, and the comparison is unblinded. These overlap percentages are therefore not independently checkable and may be optimistic. This is a load-bearing issue. I ask the authors to release the matching data and to report either dual independent annotation with agreement statistics or a transparent, rule-based matching procedure. In addition, the six compared texts are all canonical works likely present in LLM training data; the s
  2. [Sec. 4.2.3, Scene Boundary Analysis] The scene-boundary analysis labels each of the n = 3796 boundaries by 'examining the explanation provided by the LLM when starting a new scene and grouping these thematically.' This measures the LLM's stated rationale, not whether the boundary is correct. Because the LLM can rationalize an incorrect scene division, the percentages in Fig. 3B (72.7% location-related, etc.) do not validate the quality of scene segmentation; they only characterize the model's self-reported reasoning. The paper also does not report agreement between LLM scene divisions and human annotations of scene boundaries. Given that scene data is the foundation of the ribbon visualization, the authors should either reframe this analysis as an exploratory taxonomy of LLM explanations or add an independent human-annotation comparison to support the assumption that the extracted scenes are meaningful.
  3. [Sec. 4.2.1, Quote Accuracy] The exact-string-match quote check is a welcome objective measure, but the reported mean character-quote accuracy of 0.79 means that roughly one in five character quotes is replaced by a brief LLM explanation. The paper does not report how often this replacement occurs per story or how the displayed explanations are generated, nor does it evaluate whether the replacement explanations are faithful. Since the interface presents both quotes and explanations as evidence (e.g., Fig. 6B), the trust-calibration argument in Secs. 5.2 and 7.3.3 would be strengthened by reporting the proportion of displayed evidence items that are verbatim quotes versus generated explanations, and by assessing the correctness of the latter.
  4. [Sec. 4.2.1, Performance comparisons] The text states that 'there were no significant performance differences based on story length, or between well- and lesser-known texts,' but no statistical test, effect size, or confidence interval is reported. This claim is used to support the pipeline's generality. Please provide the actual test results (e.g., correlation coefficient, t-test or ANOVA output, or a nonparametric equivalent) or soften the claim to a descriptive observation.
minor comments (4)
  1. [Sec. 4.2.2, Results] The statements that '42.6% of total characters and 91.3% of themes were only detected by the LLM' should be interpreted carefully: study guides are curated and intentionally selective, so one-sided 'extra' detections are not necessarily a quality signal. The authors acknowledge this in part, but the presentation would benefit from a clearer distinction between coverage and precision.
  2. [Sec. 5.4, Implementation] The paper states that 'Our LLM prompts and example user queries are included in the supplementary materials,' but no supplementary file is visible in the submitted version. Prompts are central to reproducibility in LLM-based systems; please make them available in the final version.
  3. [Fig. 3 and Table 2] Figure 3A reports means and 95% CIs but does not state the number of stories in each category (e.g., plays vs. non-plays) or the statistical method behind 'no significant differences.' A small note in the caption would help.
  4. [Sec. 7.3.1, Context and granularity challenges] The examples of Polyphemus/Cyclops and 'Gregor's parents' are useful concrete limitations, but they also suggest that the duplicate-grouping correction loop fails in cases where entities are semantically related but not identical. A sentence connecting these user-observed failures to the quantitative limits of the correction loop would sharpen the reliability discussion.

Circularity Check

1 steps flagged · score 2.0 of 10

No central circularity: the reliability claim rests on external study-guide comparisons and exact string matches; only a minor self-confirming scene-boundary analysis and a non-load-bearing self-citation keep the score at 2.

  1. self definitional [Sec. 4.2.3, Fig. 3B; cf. Sec. 4.1.1 Step 2]
    "defining scenes based on changes in story location felt most sensible, which we implemented in our final pipeline. ... Each boundary was labeled by examining the explanation provided by the LLM when starting a new scene and grouping these thematically. ... the most common type was location change, as expected, making up 72.7% of all scene boundaries."

    The pipeline explicitly instructs the LLM to treat a change in story location as the basis for a new scene, and the evaluation then labels each boundary by reading the LLM's own explanation of why it started a new scene. Reporting that 72.7% of boundaries are location changes is therefore partly a restatement of the prompt plus the model's self-report, not an independent measure of the LLM's scene concept. The paper does acknowledge deviations (character change, focus shift, time change), and this finding is not load-bearing for the central reliability claim, so the circularity is localized and minor.

full rationale

The central derivation is self-contained. The pipeline's quantitative reliability is supported by external anchors: exact string-match quote accuracy against the raw text (Sec. 4.2.1), overlap with human-written SparkNotes/LitCharts study guides (Sec. 4.2.2), and user studies. None of these reduce by construction to fitted parameters or to self-authored results. The only self-citation found is reference [62] (Van Ham, Wattenberg, and Viégas, 'Mapping text with phrase nets'), cited in related work for exploring dependency links; it is not load-bearing. The one localized self-confirming evaluation is the scene-boundary classification in Sec. 4.2.3, where the LLM's own explanations are used as labels for a definition the pipeline itself supplied. Since that analysis does not support the central claim that the pipeline is 'sufficiently reliable to be helpful to users,' the score remains low. Concerns about unblinded qualitative matching in the study-guide comparison and possible memorization of canonical texts are validity threats, not circularity: the comparison is still made against external human-written guides, and the paper separately reports exact string-match quote accuracy and includes LLM-generated synthetic novels as a training-data control.

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

The central claim does not rest on fitted numerical parameters or novel entities. It rests on domain assumptions about scene definitions, the reliability of the second LLM for deduplication, and the validity of study guides as ground truth. The scene-boundary evaluation also assumes the LLM's explanations are faithful enough to label boundary types.

assumptions (4)
  • domain assumption Scenes in a novel are best operationalized as changes in story location.
    The pipeline prompts the LLM to define scenes by location changes (Sec. 4.1.1), a choice made by the authors after consulting literary scholars. The core data structure of the whole system depends on this.
  • domain assumption A second LLM (claude-3-5-sonnet) can reliably group duplicate character/location names and character groups without independent verification.
    The 'Correction Loop: Group Duplicate Elements' in Sec. 4.1.1 trusts the second LLM's grouping; no independent evaluation of grouping quality is reported.
  • domain assumption LLM explanations for scene boundaries are faithful enough to classify boundary types.
    In Sec. 4.2.3, each of 3796 scene divisions is labeled by examining the LLM's explanation of why a new scene started. If the LLM rationalizes its own splits post hoc, the classification is partially self-confirming.
  • domain assumption SparkNotes and LitCharts provide a reasonable ground-truth proxy for characters, themes, and events in well-known works.
    Sec. 4.2.2 uses these study guides as the only external benchmark; their own subjectivity is acknowledged (36% overlap between the two guides), so they are a weak gold standard.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Story Ribbons: Reimagining Storyline Visualizations with Large Language Models." pith.science (2026). https://pith.science/paper/E74WJNXC

@misc{pith2026250806772,
  author       = {Pith},
  title        = {Pith review of: Story Ribbons: Reimagining Storyline Visualizations with Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E74WJNXC}},
  note         = {Machine review of arXiv:2508.06772}
}
read the original abstract

Analyzing literature involves tracking interactions between characters, locations, and themes. Visualization has the potential to facilitate the mapping and analysis of these complex relationships, but capturing structured information from unstructured story data remains a challenge. As large language models (LLMs) continue to advance, we see an opportunity to use their text processing and analysis capabilities to augment and reimagine existing storyline visualization techniques. Toward this goal, we introduce an LLM-driven data parsing pipeline that automatically extracts relevant narrative information from novels and scripts. We then apply this pipeline to create Story Ribbons, an interactive visualization system that helps novice and expert literary analysts explore detailed character and theme trajectories at multiple narrative levels. Through pipeline evaluations and user studies with Story Ribbons on 36 literary works, we demonstrate the potential of LLMs to streamline narrative visualization creation and reveal new insights about familiar stories. We also describe current limitations of AI-based systems, and interaction motifs designed to address these issues.

Figures

Figures reproduced from arXiv: 2508.06772 by the authors.

Figure 1
Figure 1. STORY RIBBONS is an interactive narrative analysis tool that visualizes LLM-extracted insights about literary works. Here, a partial visualization of Pride and Prejudice by Jane Austen is shown. Each “ribbon” represents a different character (e.g., the top pink ribbon = Elizabeth Bennet), and can be used to track interactions across novel chapters (x-axis) and locations (y-axis). Chapter titles are colored by sentim… view at source ↗
Figure 2
Figure 2. Overview of our story analysis pipeline, which is organized into a data decomposition and aggregation phase. Steps involving an LLM are [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. (A) Comparing the mean number of lines extracted per scene (left) and percentage of real quotes (right) identified across different story & analysis types. Error bars indicate 95% CIs. (B) LLM classifications of scene divisions (n = 3796) by story type [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Visualizing the overlap of key characters, themes, and events extracted by our pipeline vs. literary study guides for n = 6 stories. LLM when starting a new scene and grouping these thematically. Results. As shown in Fig. 3B, our pipeline extracted 5 main scene divisio…
Figure 6
Figure 6. Figure 6: (A) When the user clicks on a chapter, the Detail Overlay opens on the right, revealing additional details and a network visualization of character interactions. (B) Users can also toggle the main plot to view scenes from this chapter and hover for character details. •…
Figure 7
Figure 7. Figure 7: Viewing raw chapter text inside of the Detail Overlay. Here, users can ask the LLM a question about the current scene or chapter. colors are chosen from a discrete color scale, while importance uses a continuous, sequential scale and sentiment uses a continuous, diverg…
Figure 8
Figure 8. Figure 8: Users can customize (A) character ribbons and (B) chapter labels in the Settings Sidebar. (C) Hovering on a character highlights their ribbon in the main plot, along with the corresponding overlay. Our backend uses Langchain and gpt-4o-mini to power all on￾the-fly feat…
Figure 9
Figure 9. Figure 9: STORY RIBBONS visualization of scenes in The Metamorphosis. (A) The default view with locations along the y-axis. (B) After ranking characters by “hope” and coloring the ribbons by “sense of duty.” Corresponding color explanations for Gregor and Grete are included. 7.2…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

78 extracted references · 39 canonical work pages

  1. [1]

    Antol, A

    S. Antol, A. Agrawal, J. Lu, M. Mitchell, D. Batra, C. L. Zitnick, and D. Parikh. Vqa: Visual question answering. In Proceedings of the IEEE international conference on computer vision, pp. 2425–2433, 2015. doi: 10.1109/ICCV.2015.279 9

  2. [2]

    Arendt and M

    D. Arendt and M. Pirrung. The “y” of it matters, even for storyline visualization. In 2017 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 81–91, 2017. doi: 10.1109/V AST.2017.8585487 2, 5

  3. [3]

    Bernstein

    M. Bernstein. Patterns of hypertext. In Proceedings of the ninth ACM conference on Hypertext and hypermedia: links, objects, time and space, pp. 21–29, 1998. doi: 10.1145/276627.276630 2

  4. [4]

    M. S. Bernstein, G. Little, R. C. Miller, B. Hartmann, M. S. Ackerman, D. R. Karger, D. Crowell, and K. Panovich. Soylent: a word processor with a crowd inside. In Proceedings of the 23nd annual ACM symposium on User interface software and technology, pp. 313–322, 2010. doi: 10. 1145/1866029.1866078 3, 4

  5. [5]

    I. Burkett. An introduction to co-design. Sydney: Knode, 12:12, 2012. 2

  6. [6]

    K. E. Chu, P. Keikhosrokiani, and M. P. Asl. A topic modeling and sentiment analysis model for detection and visualization of themes in literary texts. Pertanika Journal of Science & Technology, 30(4):2535– 2561, 2022. doi: 10.47836/pjst.30.4.14 1, 3

  7. [7]

    J. J. Y . Chung, W. Kim, K. M. Yoo, H. Lee, E. Adar, and M. Chang. Tale- brush: Sketching stories with generative pretrained language models. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems, pp. 1–19, 2022. doi: 10.1145/3491102.3501819 9

  8. [8]

    J. J. Y . Chung and M. Kreminski. Patchview: Llm-powered worldbuilding with generative dust and magnet visualization. In Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, pp. 1–19, 2024. doi: 10.1145/3654777.3676352 9

Show all 78 references
  1. [9]

    Costa and S

    M. Costa and S. Nunes. Newslines: Narrative visualization of news stories. In Text2Story@ ECIR, pp. 37–46, 2023. 2

  2. [10]

    Daniel, P

    F. Daniel, P. Kucherbaev, C. Cappiello, B. Benatallah, and M. Allahbakhsh. Quality control in crowdsourcing: A survey of quality attributes, assess- ment techniques, and assurance actions. ACM Computing Surveys (CSUR), 51(1):1–40, 2018. doi: 10.1145/3148148 3

  3. [12]

    Di Giacomo, W

    E. Di Giacomo, W. Didimo, G. Liotta, F. Montecchiani, and A. Tap- pini. Storyline visualizations with ubiquitous actors. In International Symposium on Graph Drawing and Network Visualization, pp. 324–332. Springer, 2020. doi: 10.1007/978-3-030-68766-3_25 2

  4. [13]

    V . Dibia. Lida: A tool for automatic generation of grammar-agnostic visualizations and infographics using large language models. In Proceed- ings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), pp. 113–126, 2023. ...

  5. [14]

    Dobson, P

    T. Dobson, P. Michura, S. Ruecker, M. Brown, and O. Rodriguez. Inter- active visualizations of plot in fiction. Visible Language, 45(3):169–191,

  6. [15]

    K. Elkins. The shapes of stories: sentiment analysis for narrative. Cam- bridge University Press, 2022. doi: 10.1017/9781009270403 3

  7. [16]

    Frermann, J

    L. Frermann, J. Li, S. Khanehzar, and G. Mikolajczak. Conflicts, villains, resolutions: Towards models of narrative media framing. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 8712–8732, 2023. doi: 10.18...

  8. [17]

    G. Freytag. Technique of the drama: An exposition of dramatic composi- tion and art. S. Griggs, 1895. 2

  9. [18]

    Grunde-McLaughlin, M

    M. Grunde-McLaughlin, M. S. Lam, R. Krishna, D. S. Weld, and J. Heer. Designing llm chains by adapting techniques from crowdsourcing work- flows. ACM Trans. Comput.-Hum. Interact., 2025. doi: 10.1145/3716134 3, 4

  10. [19]

    Hassan-Montero and V

    Y . Hassan-Montero and V . Herrero-Solana. Improving tag-clouds as visual information retrieval interfaces. arXiv preprint arXiv:2401.04947, 2024. doi: 10.48550/arXiv.2401.04947 1

  11. [20]

    M. N. Hoque, B. Ghai, K. Kraus, and N. Elmqvist. Portrayal: Leveraging nlp and visualization for analyzing fictional characters. In Proceedings of the 2023 ACM Designing Interactive Systems Conference, pp. 74–94,

  12. [21]

    House and A

    N. House and A. Johnston. Clover connections: Visualising character dynamics in novels for non-experts. In Proceedings of the 35th Australian Computer-Human Interaction Conference, pp. 191–201, 2023. doi: 10. 1145/3638380.3638384 1, 2, 3

  13. [22]

    Huang, J

    Y . Huang, J. Song, Z. Wang, S. Zhao, H. Chen, F. Juefei-Xu, and L. Ma. Look before you leap: An exploratory study of uncertainty measurement for large language models. IEEE Trans. Softw. Eng., 51(2):413–429, 17 pages, 2025. doi: 10.1109/TSE.2024.3519464 9

  14. [23]

    Hulstein, V

    G. Hulstein, V . Peña-Araya, and A. Bezerianos. Geo-storylines: Integrat- ing maps into storyline visualizations. IEEE Transactions on Visualization and Computer Graphics , 29(1):994–1004, 2022. doi: 10.1109/TVCG. 2022.3209480 2

  15. [24]

    Jaipersaud, Z

    B. Jaipersaud, Z. Zhu, F. Rudzicz, and E. Creager. Show, don’t tell: Uncovering implicit character portrayal using llms. arXiv preprint arXiv:2412.04576, 2024. doi: 10.48550/arXiv.2412.04576 1, 2

  16. [25]

    M. John, M. Baumann, D. Schuetz, S. Koch, and T. Ertl. A visual approach for the comparative analysis of character networks in narrative texts. In IEEE Pacific Visualization Symposium (PacificVis), pp. 247–256, 2019. doi: 10.1109/PacificVis.2019.00037 6

  17. [26]

    Kaser and D

    O. Kaser and D. Lemire. Tag-cloud drawing: Algorithms for cloud visual- ization. arXiv preprint cs/0703109, 2007. doi: 10.48550/arXiv.cs/0703109 1

  18. [27]

    D. A. Keim and D. Oelke. Literature fingerprinting: A new method for visual literary analysis. In IEEE Symposium on Visual Analytics Science and Technology, pp. 115–122, 2007. doi: 10.1109/V AST.2007.4389004 2

  19. [28]

    N. W. Kim, B. Bach, H. Im, S. Schriber, M. Gross, and H. Pfister. Vi- sualizing nonlinear narratives with story curves. IEEE transactions on visualization and computer graphics, 24(1):595–604, 2017. doi: 10.1109/ TVCG.2017.2744118 1, 2, 5

  20. [29]

    Y . Kim, Y . Chang, M. Karpinska, A. Garimella, V . Manjunatha, K. Lo, T. Goyal, and M. Iyyer. Fables: Evaluating faithfulness and content selec- tion in book-length summarization. In Proceedings of the 1st Conference on Language Modeling (COLM), 2024. doi: 10.48550/arXiv.2404...

  21. [30]

    Labatut and X

    V . Labatut and X. Bost. Extraction and analysis of fictional character networks: A survey. ACM Computing Surveys (CSUR), 52(5):1–40, 2019. doi: 10.1145/3344548 6

  22. [31]

    Lewis, E

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. Küt- tler, M. Lewis, W.-t. Yih, T. Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Proceedings of the 34th International Conference on Neural Information Processing ...

  23. [32]

    S. Liu, Y . Wu, E. Wei, M. Liu, and Y . Liu. Storyflow: Tracking the evolution of stories. IEEE Transactions on Visualization and Computer Graphics, 19(12):2436–2445, 2013. doi: 10.1109/TVCG.2013.196 1, 2, 5

  24. [33]

    J. Lu, B. Pan, J. Chen, Y . Feng, J. Hu, Y . Peng, and W. Chen. Agentlens: Visual analysis for agent behaviors in llm-based autonomous systems. IEEE Transactions on Visualization and Computer Graphics, 2024. doi: 10.1109/TVCG.2024.3394053 2

  25. [34]

    Masson, Z

    D. Masson, Z. Zhao, and F. Chevalier. Visual writing: Writing by manipu- lating visual representations of stories. arXiv preprint arXiv:2410.07486,

  26. [35]

    McCurdy, J

    N. McCurdy, J. Lein, K. Coles, and M. Meyer. Poemage: Visualizing the sonic topology of a poem. IEEE transactions on visualization and com- puter graphics, 22(1):439–448, 2015. doi: 10.1109/TVCG.2015.2467811 2, 3

  27. [36]

    Michel, E

    G. Michel, E. V . Epure, R. Hennequin, and C. Cerisara. Improving quotation attribution with fictional character embeddings. In Findings of the Association for Computational Linguistics: EMNLP 2024, pp. 12723– 12735, 2024. doi: 10.18653/v1/2024.findings-emnlp.744 4

  28. [37]

    Min and J

    S. Min and J. Park. Modeling narrative structure and dynamics with net- works, sentiment analysis, and topic modeling.PloS one, 14(12):e0226025,

  29. [38]

    F. Moretti. Distant reading. Verso Books, 2013. doi: 10.1093/llc/fqu010 9

  30. [39]

    M. E. Mortenson. Mathematics for computer graphics applications. In- dustrial Press Inc., 1999. doi: 10.5555/520335 6

  31. [40]

    R. Munroe. Movie narrative charts. https://xkcd.com/657/, Decem- ber 2009. 2, 5

  32. [41]

    Narechania, A

    A. Narechania, A. Srinivasan, and J. Stasko. Nl4dv: A toolkit for gener- ating analytic specifications for data visualization from natural language queries. IEEE Transactions on Visualization and Computer Graphics , 27(2):369–379, 2020. doi: 10.1109/TVCG.2020.3030378 2, 9

  33. [42]

    Oelke, D

    D. Oelke, D. Kokkinakis, and M. Malm. Advanced visual analytics methods for literature analysis. In Proceedings of the 6th Workshop on Language Technology for Cultural Heritage, Social Sciences, and Humanities, pp. 35–44, 2012. doi: 10.5555/2390357.2390364 2

  34. [43]

    Otake, S

    T. Otake, S. Yokoi, N. Inoue, R. Takahashi, T. Kuribayashi, and K. Inui. Modeling event salience in narratives via barthes’ cardinal functions. In Proceedings of the 28th International Conference on Computational Lin- guistics, pp. 1784–1794, 2020. doi: 10.18653/v1/2020.coling...

  35. [44]

    Padia, K

    K. Padia, K. H. Bandara, and C. G. Healey. A system for generating story- line visualizations using hierarchical task network planning. Computers & Graphics, 78:64–75, 2019. doi: 10.1016/j.cag.2018.11.004 2

  36. [45]

    Peña-Araya, T

    V . Peña-Araya, T. Xue, E. Pietriga, L. Amsaleg, and A. Bezerianos. Hy- perstorylines: Interactively untangling dynamic hypergraphs. Information Visualization, 21(1):38–62, 2022. doi: 10.1177/14738716211045007 2

  37. [46]

    Piper and S

    A. Piper and S. Bagga. Using large language models for understanding narrative discourse. In Proceedings of the The 6th Workshop on Narrative Understanding, pp. 37–46, 2024. doi: 10.18653/v1/2024.wnu-1.4 2

  38. [47]

    Piper, M

    A. Piper, M. Xu, and D. Ruths. The social lives of literary characters: Combining citizen science and language models to understand narrative social networks. In Proceedings of the 4th International Conference on Natural Language Processing for Digital Humanities, pp. 472–482,...

  39. [48]

    Qiang and C

    L. Qiang and C. Bingjie. Storycake: A hierarchical plot visualization method for storytelling in polar coordinates. In International Conference on Cyberworlds (CW), pp. 211–218. IEEE, 2016. doi: 10.1109/CW.2016. 43 2

  40. [49]

    Rashkin, A

    H. Rashkin, A. Celikyilmaz, Y . Choi, and J. Gao. Plotmachines: Outline- conditioned generation with dynamic plot state tracking. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 4274–4295, 2020. doi: 10.18653/v1/2020.emnlp-main.349 4

  41. [50]

    Rebora et al

    S. Rebora et al. Sentiment analysis in literary studies. a critical survey. Digital Humanities Quarterly, 17(2):1–17, 2023. 3

  42. [51]

    Schwan, J

    H. Schwan, J. Jacke, R. Kleymann, J.-E. Stange, and M. Dörk. Narrela- tions—visualizing narrative levels and their correlations with temporal phenomena. DHQ: Digital Humanities Quarterly, 13(3), 2019. 2

  43. [52]

    Scrivner and J

    O. Scrivner and J. Davis. Interactive text mining suite: Data visualization for literary studies. In CDH@ TLT, pp. 29–38, 2017. 2, 3

  44. [53]

    Segel and J

    E. Segel and J. Heer. Narrative visualization: Telling stories with data. IEEE transactions on visualization and computer graphics, 16(6):1139– 1148, 2010. doi: 10.1109/TVCG.2010.179 1, 2

  45. [54]

    J. Shen, J. Mire, H. W. Park, C. Breazeal, and M. Sap. Heart-felt narra- tives: Tracing empathy and narrative style in personal stories with llms. In Proceedings of the 2024 Conference on Empirical Methods in Natu- ral Language Processing, pp. 1026–1046, 2024. doi: 10.18653/v1...

  46. [55]

    Shneiderman

    B. Shneiderman. The eyes have it: A task by data type taxonomy for information visualizations. In Proceedings IEEE Symposium on Visual Languages, pp. 336–343, 1996. doi: 10.1109/VL.1996.545307 5

  47. [56]

    J. D. Spence. The memory palace of Matteo Ricci. Penguin, 1985. doi: 10. 2307/2056104 9

  48. [57]

    M. Steen. Co-design as a process of joint inquiry and imagination. Design issues, 29(2):16–28, 2013. doi: 10.1162/DESI_a_00207 2

  49. [58]

    Tanahashi and K.-L

    Y . Tanahashi and K.-L. Ma. Design considerations for optimizing story- line visualizations. IEEE Transactions on Visualization and Computer Graphics, 18(12):2679–2688, 2012. doi: 10.1109/TVCG.2012.212 2

  50. [59]

    T. Tang, R. Li, X. Wu, S. Liu, J. Knittel, S. Koch, T. Ertl, L. Yu, P. Ren, and Y . Wu. Plotthread: Creating expressive storyline visualizations using reinforcement learning. IEEE Transactions on Visualization and Computer Graphics, 27(2):294–303, 2020. doi: 10.1109/TVCG.2020....

  51. [60]

    T. Tang, S. Rubab, J. Lai, W. Cui, L. Yu, and Y . Wu. istoryline: Effective convergence to hand-drawn storylines. IEEE transactions on visualization and computer graphics, 25(1):769–778, 2018. doi: 10.1109/TVCG.2018. 2864899 2, 5

  52. [61]

    Vaithilingam, E

    P. Vaithilingam, E. L. Glassman, J. P. Inala, and C. Wang. Dynavis: Dy- namically synthesized ui widgets for visualization editing. In Proceedings of the CHI Conference on Human Factors in Computing Systems, pp. 1–17,

  53. [62]

    Van Ham, M

    F. Van Ham, M. Wattenberg, and F. B. Viégas. Mapping text with phrase nets. IEEE transactions on visualization and computer graph- ics, 15(6):1169–1176, 2009. doi: 10.1109/TVCG.2009.165 2

  54. [63]

    Venkatraman, N

    S. Venkatraman, N. I. Tripto, and D. Lee. Collabstory: Multi-llm col- laborative story generation and authorship analysis. In Findings of the Association for Computational Linguistics: NAACL 2025, pp. 3665–3679,

  55. [64]

    T. Wang, J. He, and C. Xiong. Ragviz: Diagnose and visualize retrieval- augmented generation. In Proceedings of the 2024 Conference on Empir- ical Methods in Natural Language Processing: System Demonstrations, pp. 320–327, 2024. doi: 10.18653/v1/2024.emnlp-demo.33 9

  56. [65]

    doi: 10.1145/3613904.3642639 2, 9

  57. [66]

    Watson, S

    K. Watson, S. S. Sohn, S. Schriber, M. Gross, C. M. Muniz, and M. Ka- padia. Storyprint: An interactive visualization of stories. In Proceedings of the 24th international conference on intelligent user interfaces , pp. 303–311, 2019. doi: 10.1145/3301275.3302302 1, 2, 3, 5

  58. [67]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. In Proceedings of the 36th International Conference on Neural Information Processing Systems, article no. 1800, 14 pages,...

  59. [68]

    A. Wu, Y . Wang, X. Shu, D. Moritz, W. Cui, H. Zhang, D. Zhang, and H. Qu. Ai4vis: Survey on artificial intelligence approaches for data visualization. IEEE Transactions on Visualization and Computer Graphics, 28(12):5049–5070, 2021. doi: 10.1109/TVCG.2021.3099002 2

  60. [69]

    W. Yang, M. Liu, Z. Wang, and S. Liu. Foundation models meet visual- izations: Challenges and opportunities. Computational Visual Media, pp. 1–26, 2024. doi: 10.1007/s41095-023-0393-x 2

  61. [70]

    Ware and R

    S. Ware and R. Young. Modeling narrative conflict to generate interesting stories. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, vol. 6, pp. 210–215, 2010. doi: 10. 1609/aiide.v6i1.12411 3

  62. [71]

    L. Ye, L. Wang, S. Ruan, Y . Meng, Y . Wang, W. Chen, and Z. Zhou. Storyexplorer: A visualization framework for storyline generation of textual narratives. arXiv preprint arXiv:2411.05435, 2024. doi: 10.48550/ arXiv.2411.05435 1, 2

  63. [72]

    Zamenopoulos and K

    T. Zamenopoulos and K. Alexiou. Co-design as collaborative research. Bristol University/AHRC Connected Communities Programme, 2018. 2

  64. [73]

    A. Zehe, L. Konle, L. K. Dümpelmann, E. Gius, A. Hotho, F. Jannidis, L. Kaufmann, M. Krug, F. Puppe, N. Reiter, et al. Detecting scenes in fiction: A new segmentation task. In Proceedings of the 16th conference of the European chapter of the association for computational lingu...

  65. [74]

    Zhang, M

    X. Zhang, M. Chen, and J. May. Salience-aware event chain modeling for narrative understanding. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 1418–1428, 2021. doi: 10.18653/v1/2021.emnlp-main.107 3

  66. [75]

    Z. Yang, Z. Liu, J. Zhang, C. Lu, J. Tai, T. Zhong, Y . Li, S. Zhao, T. Yao, Q. Liu, et al. Analyzing nobel prize literature with large language models. arXiv preprint arXiv:2410.18142, 2024. doi: 10.48550/arXiv.2410.18142 9

  67. [2019]

    doi: 10.1371/journal.pone.0226025 1, 2, 3, 6

  68. [2023]

    doi: 10.1145/3563657.3596000 1, 2, 3

  69. [2024]

    doi: 10.48550/arXiv.2410.07486 9

  70. [2025]

    doi: 10.18653/v1/2025.findings-naacl.203 7

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.