IndoVSE: Indonesian News Video Named Entities with Salience Labels

IndoVSE-dataset is the first Indonesian benchmark dataset for named entity salience detection in news videos. It contains 27,115 entity instances extracted from 573 unique videos, with 2,193 (8.1%) annotated as salient. The dataset was created by galihkjaya and last updated on Hugging Face in May 2026.

Use Cases

• Training models for named entity salience detection based on binary salience labels.
• Analyzing temporal patterns of entity mentions in videos based on word-level timestamps.
• Benchmarking performance of information extraction systems on Indonesian news video content.
• Studying the distribution and prominence of entities in a specific media domain.

Strengths

• First benchmark dataset for its specific task in the Indonesian language.
• Contains 27,115 annotated entity instances across 573 videos.
• Provides binary salience labels and temporal features derived from word-level timestamps.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The description metadata is limited; actual data quality requires manual inspection after download.
• Data may reflect geographic and media source bias inherent to its collection from Indonesian news videos.

Provenance

Source

huggingface

Collection Method

Extracted and annotated from Indonesian news videos.

Freshness

Last updated 2026-05-23 03:35:58; freshness should be verified.

Geography

Indonesia