StoryExplorer: A Visualization Framework for Storyline Generation of Textual Narratives

TL;DR

The result shows that users can better extract the storyline by using StoryExplorer along with the proposed workflow, which combines stroke annotation and GPT-based visual hints to quickly extract story fragments and interactively construct storyline.

Abstract

In the context of the exponentially increasing volume of narrative texts such as novels and news, readers struggle to extract and consistently remember storyline from these intricate texts due to the constraints of human working memory and attention span. To tackle this issue, we propose a visualization approach StoryExplorer, which facilitates the process of knowledge externalization of narrative texts and further makes the form of mental models more coherent. Through the formative study and close collaboration with 2 domain experts, we identified key challenges for the extraction of the storyline. Guided by the distilled requirements, we then propose a set of workflow (i.e., insight finding-scripting-storytelling) to enable users to interactively generate fragments of narrative structures. We then propose a visualization system StoryExplorer which combines stroke annotation and GPT-based visual hints to quickly extract story fragments and interactively construct storyline. To evaluate the effectiveness and usefulness of StoryExplorer, we conducted 2 case studies and in-depth user interviews with 16 target users. The result shows that users can better extract the storyline by using StoryExplorer along with the proposed workflow.

Figures (7)

• Figure 1: The workflow of narrative text knowledge externalization using StoryExplorer. (b)StoryExplorer allows readers to highlight key sentences in the text, and then StoryExplorer recognizes entities using GPT-based visual hints, followed by the selection of entities by stroke annotation. (c)Readers can organize these entities, and then add additional information, i.e., keywords and event summarization, to form fragments. (d) Readers can further explore and modify the storyline to create a unified storytelling representation.
• Figure 2: Formative study result: Participants annotated raw text from (a) entity books and (b) Adobe Acrobat to form working memory and ultimately conveyed the working memory using (c) storyline to promote long-term memory.
• Figure 3: StoryExplorer system architecture: (a) text view integrates stroke annotation and GPT-based visual hints to extract entities, (b) fragment view records user operations and generates text summarization and word cloud. (c) storyline view creates storytelling specifications, and (d) SVG renderer updates the diagram canvas.
• Figure 4: Two cases of using StoryExplorer to construct storylines for the novel "The Little Match Girl" and the script of "Schindler's List" are presented. The entire workflow follows the selection-organization-storytelling method. In Figure F5, U4 employed the highlight-select method, initially constructing a fragment and filtering characters by deleting them from the text. Figure F11 demonstrates how U4 adjusted a fragment that was misclassified as occurring at the same time. Figure F13 illustrates the interconnectivity of all views, where actions in the fragment view are reflected in the text view. Figure b1 displays the relatively flat storyline constructed from the script. In Figure b2, crucial time points are highlighted in the storyline. In Figure b3, two events at the same location are emphasized, indicating their corresponding places. Figure b4 showcases the script's construction of storylines with multiple single-character fragments.
• Figure 5: User interview Results: (a) the first seven fragments of each operation time statistics for Task 1 “Harry Potter and the Sorcerer's Stone first three chapters”, where the block size indicates the length of time spent. (b) StoryExplorer log-data visualization showing different actions. (c) Storyline created by U4 for task 1 "Harry Potter and the Sorcerer's Stone first three chapters" and U8 for task 2 "Pride & Prejudice". (d) Similarity of final storyline and completion time statistics for both tasks. As the complexity of the text increases, both the time costs and content similarity are significantly affected.