Source-linked AI summary
Text-to-Viz: Automatic Generation of Infographics from Proportion-Related Natural Language Statements
Weiwei Cui, Xiaoyu Zhang, Yun Wang, He Huang, Bei Chen, Lei Fang, Haidong Zhang, Jian-Guan Lou, Dongmei Zhang
TL;DR
Professional infographic creation is time-consuming and often requires design expertise, limiting the appeal of existing authoring tools for casual users. The paper introduces a proof-of-concept system that converts proportion-related natural-language statements into pre-designed infographics, with sample results, exhibits, and expert reviews demonstrating usability and promise. Its scope remains limited to a relatively small information space and template-based styles, with failures also arising from text analysis and icon selection.
Problem
Creating professional infographics with existing tools requires substantial time and design expertise, making them less attractive to casual users.
Method
The system analyzes infographic design patterns and converts simple proportion-related natural-language statements into sets of infographics with pre-designed styles.
Results
Sample results, public exhibits, and expert reviews demonstrate the system’s usability and usefulness for producing professional-looking infographic options.
Takeaways & Limitations
Casual users can select or refine generated infographic styles without a complex authoring process or design expertise.
Takeaways & Limitations
The current approach handles only a relatively small set of information, relies on pre-designed styles, and can fail through incorrect text analysis or icon matching.
Abstract
from arXiv · showhide
Combining data content with visual embellishments, infographics can effectively deliver messages in an engaging and memorable manner. Various authoring tools have been proposed to facilitate the creation of infographics. However, creating a professional infographic with these authoring tools is still not an easy task, requiring much time and design expertise. Therefore, these tools are generally not attractive to casual users, who are either unwilling to take time to learn the tools or lacking in proper design expertise to create a professional infographic. In this paper, we explore an alternative approach: to automatically generate infographics from natural language statements. We first conducted a preliminary study to explore the design space of infographics. Based on the preliminary study, we built a proof-of-concept system that automatically converts statements about simple proportion-related statistics to a set of infographics with pre-designed styles. Finally, we demonstrated the usability and usefulness of the system through sample results, exhibits, and expert reviews.
1 INTRODUCTION
Infographics combine data-driven content with visual embellishments, but creating professional results remains time-consuming and difficult for casual users. The paper proposes automatically generating proportion-focused infographics from natural-language statements using analyzed design patterns and a synthesis framework.
- 1 INTRODUCTION: Professional infographic creation combines data content with artistic elements but requires substantial time and designer skills.
- 1 INTRODUCTION: Casual creators typically need effective, professional samples rather than highly creative designs and can select quickly when relevant examples are provided.
- 1 INTRODUCTION: Automatic generation from natural-language statements targets casual users who lack design expertise or time for complex infographic authoring.The approach addresses both tool unfamiliarity and the difficulty of conceiving a professional design from scratch.
- 1 INTRODUCTION: The system addresses text understanding and infographic construction by extracting information with a CRF-based model and synthesizing designs from analyzed exemplars.
- 1 INTRODUCTION: The proof-of-concept focuses on proportion facts and converts simple statements into infographics with pre-designed styles, evaluated through samples, exhibits, and expert reviews.
2 RELATED WORK
Prior systems support infographic design, chart generation, and natural-language data analysis, but generally require structured data, technical concepts, or an initial design idea. This work instead converts explicitly stated information directly into infographic-style visualizations for users without design experience.
- 2 RELATED WORK: Natural-language visualization tools typically translate user queries into database queries, while this system directly visualizes the user’s input.
- 2 RELATED WORK: Automatic chart-generation systems reduce encoding effort but require specific data formats and often target professional users exploring complex patterns.
- 2 RELATED WORK: Existing visualization systems generate charts or analyze datasets, whereas this system directly converts explicitly provided information into infographic-style visualizations.
- 2 RELATED WORK: Interactive infographic authoring tools offer flexible manipulation but require concepts such as vectors, layers, and data binding, plus an initial design idea.
- 2 RELATED WORK: The proposed system trades flexibility for reduced authoring complexity to help users without design experience create infographics automatically.
3 PRELIMINARY SURVEY OF INFOGRAPHICS
The survey decomposed collected infographics into valid units and characterized their information types, finding recurring statistical subcategories and relationships among multiple facts.
- Corpus construction: The corpus was built from the first 200 distinct Google Image Search results and decomposed into valid infographic units meeting message, graphical-element, coherence, and indivisibility criteria.The resulting percentages are based on 983 infographic units rather than 200 infographic sheets.
- Information categories: Statistical-based infographics were categorized by underlying message into proportion, quantity, change, and rank subcategories.Proportion was selected for the prototype because it was common and had distinctive visual and textual patterns.
- Visual patterns: The survey examples show recurring visual strategies for quantity, change, and rank, including icons or charts, contrast colors with comparisons or arrows, and highlighted or ordered displays.The examples organize visual treatments by the underlying statistical message.
4 PROPORTION-RELATED INFORMATION
The paper focuses on proportion-related statements and analyzes their textual descriptions, visual layouts, graphics, and color rules as a basis for automatic infographic generation.
- Scope: Proportion-related facts describe how much a part occupies the whole and commonly use forms such as n%, m in n, m/n, or “half of.”The authors target this category because it was prevalent in the collected data and exhibited prominent visual and textual patterns.
- Design space: The proportion design space has four dimensions: layout, description, graphic, and color, identified through discussion with two experienced graphic designers.The discussion used examples of proportion-related infographics and examined how designers generate them from scratch.
- Layout: Single-fact proportion infographics typically combine descriptions and graphics in horizontal, vertical, tiled, or overlaid arrangements.Overlaying is especially used when geographic information is shown on a map.
- Description: Descriptions may include the entire statement, a number-removed statement, a verb-object phrase, and the number itself.These text blocks can be combined to communicate the complete proportion message.
- Graphic: Graphics should relate semantically to the statement, with common choices including pictographs, adornments, donut and pie charts, bar charts, filled icons, and scaled icons.Different graphic elements can serve different communicative roles.
- Color: Color design uses contrasting background and foreground colors, while graphic elements may use one or more colors according to their roles.Color themes may also indicate latent semantics, such as environmental associations with green or blue.
5 TEXT-TO-VIZ IMPLEMENTATION
The proof-of-concept system converts a user’s proportion statement into an infographic by combining text analysis with visual generation.
- System architecture: The system has two modules: a text analyzer extracts statement segments, and a visual generator uses them to construct an infographic.The input is a natural-language proportion statement such as “More than 40% of students like football.”
5.1 Text Analyzer
The text analyzer performs named entity recognition for proportion statements, extracting modifiers, numbers, parts, and wholes with a supervised CNN+CRF model.
- Entity extraction: The analyzer extracts four predefined entity types—modifier (M), number (N), part (P), and whole (W)—from proportion statements.This extraction is framed as a named entity recognition task.
- Model pipeline: The model tokenizes statements, concatenates word, syntactic, and Brown-clustering features, and feeds the resulting feature matrix to a CNN followed by a CRF layer.The CNN learns token representations while the CRF determines entity labels; the parameters are jointly trained.
- Entity labeling: Figure 7 illustrates mapping statement tokens to IOB labels, where B-, I-, and O denote begin, inside, and outside positions.The labeling sequence represents entity spans in the input statement.
- Training and evaluation: The prototype trains its CNN+CRF text analyzer on 800 manually annotated statements and evaluates it with 10-fold cross-validation.The training-results table is identified as Table 2, but its metric values are not supplied here.
5.2 Visual Generator
The visual generator constructs candidate infographics by combining layout blueprints with descriptions, graphics, and color palettes, then synthesizes and ranks valid results. Its scoring evaluates semantic matching, visual fit, and information completeness, while users can replace selected elements.
- The generator enumerates layout, description, graphic, and color candidates, synthesizes valid combinations, and ranks the resulting infographics.Invalid layouts are filtered before required elements are positioned and optimized.
- 5.2.1 Layout: Layout blueprints specify aspect ratio, spatial regions, content assignments, and constraints, with 20 initial blueprints serving as an expandable pool.One example uses a 2 × 1 aspect ratio and recursively divides the canvas into three regions.
- 5.2.6 Ranking: The ranking stage scores semantic relevance, visual appearance, and informative completeness to assess message delivery quality.Semantic scoring uses Word2Vec matching, visual scoring measures wasted space, and informative scoring checks whether input words appear in the result.
- 5.2.6 Ranking: The total score combines semantic, visual, and informative scores with default weights of 0.25, 0.5, and 0.25, respectively.
- Users can replace recommended icons, color palettes, and descriptions before saving an infographic or template for reuse.
6 EVALUATION
The evaluation presented diverse generated infographics, gathered feedback from more than 80 exhibit visitors, and interviewed three professional designers. Both audiences viewed the system as useful and promising, while designers judged the output adequate for its target users and identified specific quality issues.
- 6.1 Sample Infographics: The system generated diverse designs from proportion statements, including pictographs, filled icons, tilted layouts, and semantic color choices.
- 6.2 Casual User Interview in Exhibits: More than 80 exhibit visitors gave overwhelmingly positive feedback and saw convenience for reports, presentations, and other communication tasks.
- 6.2 Casual User Interview in Exhibits: Word2Vec-based icon matching sometimes produced inappropriate recommendations, such as a skirt icon for the statement that secretaries wear glasses.
- 6.3 Expert Interview: Designers found the generated infographics impressive, usable for posters, slides, and reports, and worth using when the system became available.
- 6.3 Expert Interview: Designers appreciated text input and candidate designs as a convenient workflow and felt additional editing functions were unnecessary for the target use case.
- 6.3 Expert Interview: Designers considered the output good enough for its target users, but suggested improving typography, semantic icon meaning, and visualization accuracy.
7 DISCUSSION
The discussion positions automatic infographic generation as a more suitable paradigm for casual users while identifying limitations in usability, expressiveness, supported information, and text interpretation.
- 7 DISCUSSION: Manual editing tools are unsuitable for casual users because they assume design expertise and require users to envision the desired infographic.The proposed alternative aims to reduce interaction and design demands for occasional creators.
- 7 DISCUSSION: The proof-of-concept remains interruptive, requiring users to open the system, enter a statement, and insert the generated infographic into their work.The authors suggest integrating background text analysis into productivity software as a more seamless scenario.
- 7 DISCUSSION: Failure cases arise from incorrect text analysis, mismatched icons, awkward text wrapping, and unresolved aesthetic considerations.Long or complicated statements can produce incorrect tags, while long descriptions may force small and illegible fonts.
- 7 DISCUSSION: The approach currently handles only a small information space, relies on manually specified representations, and has uncertain extension beyond proportion facts.The discussion identifies broader statistical and infographic types as future extensions.
- 7 DISCUSSION: Pre-designed styles limit infographic expressiveness and do not provide the creativity of human designers, although users can add materials.The resulting designs remain constrained despite the framework’s openness to enrichment.
8 CONCLUSION AND FUTURE WORK
The paper concludes that its proof-of-concept can turn proportion statements into selectable professional infographics without complex authoring or design expertise. It also identifies broader information coverage and improved algorithmic quality as future directions.
- 8 CONCLUSION AND FUTURE WORK: The framework translates proportion-fact statements into multiple professional infographic styles that casual users can select or refine.Example results and user/designer interviews support the system’s usability and promise for everyday adoption.
- 8 CONCLUSION AND FUTURE WORK: Future work should support more statistical and infographic information types, while replacing rule-based selection and ranking with techniques that improve result quality.The current proof-of-concept is limited to a specific information type and uses rule-based algorithms for icon, color, and ranking decisions.