Plume: Scaffolding Text Composition in Dashboards

TL;DR

Plume tackles the lack of text authoring support in dashboards by proposing a frame-based, text-role aware system that uses large language models with human-in-the-loop oversight to generate and edit narrative text that accompanies visualizations. The authors conduct a formative study on 672 text fragments from 40 dashboards to build a granular codebook and a design space for text generation, then validate the approach with a user study of 12 dashboard authors, showing both time-saving benefits and concerns about verbosity and accuracy. The work demonstrates that structured text guidance, frame-aware placement, and role-specific generation can meaningfully augment dashboard authoring, while highlighting tradeoffs between automation and control in dynamic, data-driven contexts. It also outlines future directions, including real-time updates, multilingual support, and enhanced coherence across text roles, which could substantially impact how data stories are crafted and maintained across teams.

Abstract

Text in dashboards plays multiple critical roles, including providing context, offering insights, guiding interactions, and summarizing key information. Despite its importance, most dashboarding tools focus on visualizations and offer limited support for text authoring. To address this gap, we developed Plume, a system to help authors craft effective dashboard text. Through a formative review of exemplar dashboards, we created a typology of text parameters and articulated the relationship between visual placement and semantic connections, which informed Plume's design. Plume employs large language models (LLMs) to generate contextually appropriate content and provides guidelines for writing clear, readable text. A preliminary evaluation with 12 dashboard authors explored how assisted text authoring integrates into workflows, revealing strengths and limitations of LLM-generated text and the value of our human-in-the-loop approach. Our findings suggest opportunities to improve dashboard authoring tools by better supporting the diverse roles that text plays in conveying insights.

Figures (6)

• Figure 1: An illustration of the usage scenario. An analyst starts by laying out the dashboard and adding charts to the canvas. They may then accept Plume suggestions to fill out the initial text based on the text roles they deem necessary. The analyst then makes final edits to the text, generates more accurate text using Plume now that it is aware of the analyst's intended tone and goals, and finally uses linguistic metrics in Plume to ensure readability.
• Figure 2: The dashboards in our formative study displayed varying levels of coordination between the visual layout of text and its semantic scope. In all cases, the Gestalt principle of proximity helps trace which part of the dashboard each text describes, though with differing clarity. Dashboard (a) uses a rigid grid, with each text placed directly above the relevant chart, section, or entire dashboard. Dashboard (b) also employs a grid but sometimes places text to the side of the chart or as an annotation on top of it, still maintaining clear associations. In contrast, Dashboard (c) places most text in its own section equidistant from two views, making it harder to visually infer the corresponding view.
• Figure 3: View from composing a dashboard in Plume. When hovering over a text fragment, Plume highlights the descendant frames used for its text generation with a blue border. The hovered text remains independent of the sibling frame to the right.
• Figure 4: Snippet dropdown menu that contains snippet-level writing guidance. Based on the selected text role, Plume provides recommendations for how to approach the writing of such text as well as surfaces text metrics. The user may then either manually rewrite the text or use Plume to re-generate, shorten, or simplify their text to ensure that it is readable at the level appropriate for their audience.
• Figure 5: An illustration of the text generation logic of a section title with respect to the dashboard dependency tree.