LRAT Train: Agent Trajectory Data for Retrieval Model Training

LRAT Training Dataset contains trajectory-derived retrieval supervision for training retrievers in agentic search. The dataset is built from deep research agent trajectories that record intermediate search queries, browsing actions, and post-browse reasoning traces. It was created by Yuqi-Zhou and last updated on 2026-04 08.

Use Cases

• Training retrieval models based on agent trajectory data.
• Improving search agent performance based on query and browsing action sequences.
• Studying agent reasoning patterns based on post-browse reasoning traces.

Strengths

• Derived from deep research agent trajectories, providing a source of supervision distinct from human click logs.
• Contains intermediate search queries, browsing actions, and reasoning traces, which suggests multi-step interaction data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Built from deep research agent trajectories.

Time Range

null

Freshness

Last updated 2026-04-08 11:24:08; freshness should be verified.

Geography

null