CRAFT-WikiVideo: Multilingual News Video Chunks with ASR and Persona Queries

CRAFT-WikiVideo integrates 52 news events from the hltcoe/wikivideo dataset with processed video artifacts. The dataset includes original videos plus CRAFT pipeline outputs such as chunked video segments, multilingual automatic speech recognition transcripts, and synthesized persona queries. It was created by author mbhosale and last updated on Hugging Face in May 2026.

Use Cases

• Train video chunking models based on the adaptive keyframe selection process described.
• Develop multilingual automatic speech recognition systems using the ASR outputs included.
• Build persona-driven conversational agents based on the synthesized persona queries.
• Analyze news event narratives across modalities using the combined video and text artifacts.

Strengths

• Integrates content from 52 distinct news events, providing a foundation for event-based analysis.
• Includes outputs from a multi-stage CRAFT pipeline, offering processed data like chunked videos and ASR transcripts.
• Combines original source videos from hltcoe/wikivideo with newly generated artifacts.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

hltcoe/wikivideo dataset and outputs from the CRAFT pipeline.

Collection Method

Pipeline processing involving chunking, multilingual ASR, adaptive keyframe selection, and query synthesis.

Freshness

Last updated 2026-05-24 22:39:46; freshness should be verified.