FirstPass: Multi-Domain Scientific Peer-Review Dialogues from Nature Communications

3,668 complete peer-review dialogues across five scientific domains, built from Nature Communications transparent peer-review files. The dataset includes editorial outcome labels derived from real revision-cycle outcomes and was introduced alongside a fine-tuned model for revision-cycle prediction. It was created by Prabhjotschugh and last updated on 2026-05-30.

Use Cases

• Training models to predict manuscript revision outcomes based on peer-review dialogue content.
• Analyzing the structure and language of multi-round scientific peer-review conversations.
• Studying domain-specific patterns in reviewer feedback and editorial decisions.
• Benchmarking models for dialogue understanding and classification tasks in academic settings.

Strengths

• Contains 3,668 complete peer-review dialogues, providing a substantial corpus for analysis.
• Covers five distinct scientific domains, offering cross-disciplinary scope.
• Includes real editorial outcome labels linked to revision-cycle results.
• Associated fine-tuned model reported 80.5% accuracy and 78.2% F1-macro on a specific prediction task.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other structural details like file formats and sample data are unavailable.
• Data may reflect the specific editorial processes and biases inherent to the Nature Communications source.

Provenance

Source

Nature Communications transparent peer-review files.

Collection Method

Built from published peer-review dialogues.

Freshness

Last updated 2026-05-30 05:58:45; freshness should be verified.