DataReel Benchmark for Automated Data Video Generation
A team of researchers has unveiled DataReel, a new benchmark designed for the automated generation of data-driven video narratives, featuring 328 authentic stories. These data videos merge animated visualizations focused on charts with coordinated narration, finding applications in journalism, education, and public outreach. Although they are impactful, creating such videos demands skills in visualization design, animation, and video editing. While recent progress in large language models presents opportunities for automation, a benchmark for assessing models in animated visualization storytelling was previously lacking. DataReel addresses this need by offering a well-structured dataset for thorough evaluation. This research is documented in arXiv:2604.25220.
Key facts
- DataReel is a benchmark for automated data-driven video story generation.
- It includes 328 real-world stories.
- Data videos combine animated chart visualizations with synchronized narration.
- They are used in journalism, education, and public communication.
- Generating data videos requires expertise in visualization, animation, and video editing.
- Large language models offer new automation opportunities.
- No prior benchmark existed for animated visualization-based video storytelling.
- The research is published on arXiv with ID 2604.25220.
Entities
Institutions
- arXiv