README.md

---
license: mit
task_categories:
- question-answering
- summarization
- text-classification
language:
- en
tags:
- legal
pretty_name: D.B. Cooper Files Text
size_categories:
- 10K<n<100K
---

# D.B. Cooper FBI Files Text Dataset

This dataset contains extracted text from the FBI's case files on the infamous "DB Cooper" skyjacking (NORJAK investigation). [The files are sourced from the FBI](https://vault.fbi.gov/D-B-Cooper%20) and are provided here for open research and analysis.

## Dataset Details

- **Source:** FBI NORJAK (D.B. Cooper) case files, as released and processed in the [db-cooper-files-text project](https://github.com/noops888/db-cooper-files-text).
- **Format:** Each entry contains a chunk of extracted text, the source page, and file metadata.
- **Rows:** 44,138
- **Size:** ~63.7 MB (raw); ~26.8 MB (Parquet)
- **License:** Public domain (U.S. government work); see original repository for details.

## Motivation

This dataset was created to facilitate research and exploration of one of the most famous unsolved cases in U.S. criminal history. It enables:
- Question answering and information retrieval over the DB Cooper files.
- Text mining, entity extraction, and timeline reconstruction.
- Comparative analysis with other historical FBI files (e.g., the JFK assassination records).

## Data Structure

Each row in the dataset contains:
- `id`: Unique identifier for the text chunk.
- `content`: Raw extracted text from the FBI file.
- `sourcepage`: Reference to the original file and page.
- `sourcefile`: Name of [the original PDF file.](https://github.com/noops888/db-cooper-files-text/tree/main/download_script/pdfs)

Example:

```json
{
  "id": "file-cooper_d_b_part042_pdf-636F6F7065725F645F625F706172743034322E706466-page-5",
  "content": "The Seattle Office advised the Bureau by airtel dated 5/16/78 that approximately 80 partial latent prints were obtained from the NORJAK aircraft...",
  "sourcepage": "cooper_d_b_part042.pdf#page=4",
  "sourcefile": "cooper_d_b_part042.pdf"
}
```

## Usage

This dataset is suitable for:
- **Question answering**: Retrieve answers to questions about the DB Cooper case directly from primary sources.
- **Information retrieval**: Build search engines or retrieval-augmented generation (RAG) systems.
- **Named entity recognition**: Extract people, places, dates, and organizations from FBI documents.
- **Historical research**: Analyze investigation methods, suspects, and case developments.

## Task Categories

Besides "question answering", this dataset is well-suited for the following task categories:

- **Information Retrieval**: Document and passage retrieval from large corpora of unstructured text.
- **Named Entity Recognition (NER)**: Identifying people, places, organizations, and other entities in historical documents.
- **Summarization**: Generating summaries of lengthy case files or investigative reports.
- **Document Classification**: Categorizing documents by topic, date, or investigative lead.
- **Timeline Extraction**: Building chronological event sequences from investigative records.

## Acknowledgments

- FBI for releasing the NORJAK case files.