MetaphorVU-Bench: A Benchmark for Metaphorical Video Understanding

MetaphorVU-Bench is a benchmark dataset for metaphorical video understanding. It is characterized by a systematic taxonomy and contains metaphorical videos curated from billions of real-world candidates with rigorous human annotation. The dataset was created by author lzq2021 and was last updated on Hugging Face in May 2026.

Use Cases

• Train and evaluate multimodal AI models for metaphor detection based on curated video content.
• Benchmark video understanding systems on figurative language tasks based on the described systematic taxonomy.
• Research the intersection of computer vision and natural language semantics based on human-annotated metaphorical videos.

Strengths

• Videos are curated from billions of real-world candidates, suggesting a large-scale selection process.
• Annotations are described as 'rigorous human annotation', implying a high-quality labeling effort.
• The benchmark is built upon a 'well-founded systematic taxonomy' for metaphorical understanding.

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

huggingface

Collection Method

Curated from billions of real-world video candidates and annotated by humans.

Freshness

Last updated 2026-05-15 02:28:20; freshness should be verified.