BaixuW/aviation-domain-pretrain-corpus

Aviation Domain Pretrain Corpus

Dataset Summary

The Aviation Domain Pretrain Corpus is an early-stage domain-specific dataset designed for continued pretraining (PT) of large language models in the aviation field.

This dataset is currently in a very early stage. It is intended primarily for research exploration and method validation rather than production use.

The dataset focuses on injecting basic aviation knowledge into LLMs, including:

• —Air Traffic Control (ATC)
• —Aviation Meteorology
• —Flight Performance
• —Aeronautical Information

Intended Use

This dataset is intended for:

• —Domain adaptation of large language models
• —Preliminary research on aviation-related LLM applications
• —Supporting downstream tasks such as:
• —Flight path planning
• —Air traffic decision modeling
• —Weather-aware reasoning

Dataset Structure

The dataset consists of cleaned text samples:

{
  "text": "Example aviation-related knowledge text..."
}

Data Construction

The dataset is constructed from:

• —Aviation textbooks
• —Manuals and guidelines
• —Publicly available aviation-related materials

Processing steps include:

1. 1.Basic text cleaning
2. 2.Segmentation
3. 3.Simple structuring

Due to the early-stage nature, the processing pipeline is still evolving.

Limitations

• —Early-stage dataset with limited scale
• —Coverage is incomplete
• —Possible noise and inconsistency
• —Not suitable for safety-critical applications

Future Work

Future improvements may include:

• —Expanding dataset size
• —Improving data quality and consistency
• —Adding more structured annotations
• —Enhancing domain coverage

Citation

@dataset{aviation_domain_corpus,
  title={Aviation Domain Pretrain Corpus},
  author={BaixuW},
  year={2026}
}

Author

BaixuW