DeepDive: Training Data for Complex Multi-Step Reasoning Search Agents

The DeepDive dataset is constructed through automated knowledge graph random walks, entity obfuscation, and difficulty filtering to create challenging questions. It is designed for training deep search agents with complex, multi-step reasoning capabilities. The dataset was created by zai-org and last updated on March 17, 2026.

Use Cases

• Training search agents for complex, multi-step reasoning based on the described automated construction process.
• Benchmarking retrieval-augmented generation (RAG) systems on challenging questions created via entity obfuscation.
• Developing models that require sophisticated search skills based on the dataset's stated purpose.
• Studying automated dataset generation techniques using knowledge graph random walks.

Strengths

• Dataset is specifically designed for challenging questions requiring sophisticated search and retrieval.
• Construction method involves automated knowledge graph random walks and difficulty filtering.
• Last updated on March 17, 2026, suggesting recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown.

Provenance

Source

zai-org

Collection Method

Constructed through automated knowledge graph random walks, entity obfuscation, and difficulty filtering.

Time Range

null

Freshness

Last updated 2026-03-17 13:29:30; freshness should be verified.

Geography

null