ToolMind-Web-QA: Search-Augmented Multi-Hop Question Answering

ToolMind-Web-QA consists of 6,000 complex question-answer pairs synthesized from Wikipedia entity-relation knowledge graphs, published by Nanbeige in February 2026. It includes search trajectories averaging 100 turns per record to facilitate research on search-augmented and long-horizon agents.

Use Cases

• Training search-augmented agents using the 100-turn search trajectories
• Evaluating multi-hop reasoning capabilities via Wikipedia-derived QA pairs
• Developing evidence-grounded text generation models based on knowledge graph relations

Strengths

• 6,000 complex QA pairs
• Trajectories averaging 100 turns
• Synthesized from Wikipedia entity-relation knowledge graphs

Limitations

• Synthetic data generation may lack the nuance of human-generated search queries
• Domain coverage is restricted to Wikipedia-based entities and relations

Provenance

Source

Nanbeige (Arxiv260213367)

Collection Method

Synthetic

Freshness

Last updated February 2026.

Geography

United States