Compendia: Automated Visual Storytelling Generation From Online Article Collection

Area of Science:

• Computational Linguistics
• Data Visualization
• Computational Journalism

Background:

• Readers seek clear, quantitative journalism but often piece together scattered facts from multiple online articles.
• Visual storytelling effectively transforms fragmented information into engaging narratives, but its application to unstructured online articles is underexplored.

Purpose of the Study:

• To present Compendia, an automated system designed to generate coherent data stories from unstructured online articles tailored to user queries.
• To address challenges in analyzing unstructured text for data storytelling.

Main Methods:

• Compendia utilizes two modules: Online Article Retrieval for gathering relevant articles and Data Fact Extraction for identifying, validating, and refining quantitative facts.
• Fact Organization clusters and merges extracted facts into thematic groups, followed by Visual Storytelling to create narrative visualizations in an interactive scrollytelling interface.

Main Results:

• Quantitative analysis confirmed Compendia's accuracy in fact extraction and organization.
• Two user studies with 16 participants demonstrated the system's usability and effectiveness in producing engaging visual stories for open-ended queries.

Conclusions:

• Compendia offers a novel solution for automated data storytelling from unstructured online content.
• The system enhances user comprehension of complex topics by transforming fragmented information into coherent, visually engaging narratives.