TrialPD-Bench
A benchmark for clinical trial protocol digitization, annotated by experts. It contains 150 public protocol PDFs and covers four structured extraction tasks, and is used to evaluate the accuracy and evidence recall of coding agents in long-document conversion.
Timeline (last 90 days)
The researchers constructed TrialPD-Bench, a benchmark set annotated by experts, containing 150 public clinical trial protocol PDFs and covering four types of tasks.
The evaluation scope of TrialPD-Bench is not limited to final answer correctness, but also includes the evidence recall rate of agent trajectories, agent behavior patterns, and cost.
All Facts (2)
The researchers constructed TrialPD-Bench, a benchmark set annotated by experts, containing 150 public clinical trial protocol PDFs and covering four types of tasks.
50%UnverifiedThe evaluation scope of TrialPD-Bench is not limited to final answer correctness, but also includes the evidence recall rate of agent trajectories, agent behavior patterns, and cost.
50%