HIPPO-VIDEO: 2,040 Simulated YouTube Watch Histories for Highlighting

2,040 simulated YouTube watch histories were generated by a large language model–based user simulator for research. Each simulated history consists of 10 videos, modeling user behavior from a starting point. The dataset was proposed in a COLM 2025 paper and uploaded to Hugging Face by user jeongeunnn.

Use Cases

• Training models for video highlight prediction based on simulated watch sequences.
• Benchmarking recommendation systems using synthetic user behavior data.
• Studying the patterns of simulated watch histories generated by LLM agents.

Strengths

• Contains 2,040 distinct simulated user histories.
• Each history is structured with a sequence of 10 videos.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is simulated and may not reflect real-world user behavior patterns.

Provenance

Source

Hugging Face user jeongeunnn.

Collection Method

Generated by an LLM-based user simulator.

Freshness

Last updated 2026-05-13 08:48:36; freshness should be verified.