Show Me the Infographic I Imagine: Intent-Aware Infographic Retrieval for Authoring Support

Jing Xu, Jiarui Hu, Zhihao Shuai, Yiyun Chen, and Weikai Yang*
* Corresponding author
Paper Supplementary Video Code

Teaser

From vague intent to editable infographic drafts

Teaser overview of the retrieve-and-adapt workflow for infographic authoring.
A natural-language query is decomposed into weighted intent facets, matched against an infographic corpus, and turned into exemplar-grounded outputs that can be adapted to the user's data.

Abstract

Why retrieval needs to understand design intent

While infographics are a powerful medium for communicating data-driven stories, authoring them from scratch remains difficult, especially for novice users. Retrieving strong exemplars can reduce that burden, but effective retrieval is challenging because users often express their needs in ambiguous natural language while infographic designs blend multiple visual and structural factors. This project introduces an intent-aware infographic retrieval framework that aligns free-form requests with infographic designs through a five-facet taxonomy spanning content, style, layout, illustration, and chart type. Building on the retrieved exemplars, the system also supports exemplar-driven adaptation through a conversational interface that helps users turn high-level design goals into editable outputs.

Method

Facet-aware retrieval instead of one-score similarity

Intent taxonomy

User requests are represented through five facets so the system can separately reason about topic, chart form, composition, decorative cues, and aesthetic tone.

Weighted matching

Each facet contributes with its own query text and importance weight, which makes retrieval more controllable than a single collapsed relevance score.

Authoring support

Retrieved exemplars are not the endpoint. They become references that users can pin, compare, and adapt in an integrated authoring workflow.

Method overview showing training and inference pipelines for intent-aware infographic retrieval.
The retrieval pipeline combines taxonomy-guided query parsing, facet-specific similarity estimation, and weighted fusion over an infographic corpus.
content style layout illustration chart type

Demo

Conversational retrieve-and-adapt workflow

The video walks through retrieval, exemplar selection, and downstream adaptation inside the interactive system.

System interface with chat, exemplar panel, output panel, and retrieval panel.
The interface keeps retrieval controls, pinned exemplars, and iterative SVG outputs in one place.

Additional Views

More detail on chart types and authoring outcomes

Chart type overview figure.
Coarse chart-type organization used to support controllable matching and soft chart-type constraints.
Examples from the authoring evaluation.
Qualitative authoring cases illustrate how retrieved references support downstream adaptation.

Resources

Project files and repositories

PDF Paper

Main paper PDF for the project.

 PDF Supplementary

Additional prompts, protocol details, and system notes.

 MP4 Demo Video

Walkthrough of the retrieval-and-adaptation workflow.

 GitHub Code Repository

Separate source repository linked from this homepage.

 GitHub Homepage Repository

The GitHub Pages repository hosting this site.