jemail-house-oversight-embeddings

Dataset Description

This dataset contains text embeddings for documents from the House Oversight source related to the Jeffrey Epstein case. Each embedding is paired with its original text content, making it suitable for semantic search, document analysis, and retrieval-augmented generation (RAG) applications.

Source Information

House Oversight Committee document releases related to the Jeffrey Epstein case. These documents include emails, correspondence, and other materials released by the House Committee on Oversight and Reform.

Dataset Statistics

• Total Embeddings: 3,670
• Failed Embeddings: 10
• Success Rate: 99.73%
• Embedding Model: qwen/qwen3-embedding-8b
• Embedding Dimension: 8192 (qwen3-embedding-8b)

Dataset Structure

Each example in the dataset contains the following fields:

• doc_id (string): Unique document identifier
• source (string): Source of the document (house-oversight)
• page_number (int): Page number within the document (0-indexed for emails/messages)
• embedding (list[float]): Vector embedding of dimension 8192
• text (string): Original text content that was embedded (includes summary description + page body)
• content_length (int): Length of the embedded text in characters
• summary_length (int): Length of the summary description in characters
• page_content_length (int): Length of the page body content in characters
• truncated (bool): Whether the text was truncated before embedding
• model (string): Model used to generate the embedding
• timestamp (string): ISO timestamp when the embedding was created

Text Content Format

The text field contains the text that was embedded, which is constructed as:

{summary_description}

{page_body}

Where:

• summary_description comes from the document's summary.yaml file (if available)
• page_body is the content extracted from the markdown page file (frontmatter stripped)

Usage

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("567-labs/jemail-house-oversight-embeddings")

# Access the data
for example in dataset["train"]:
    doc_id = example["doc_id"]
    embedding = example["embedding"]
    text = example["text"]
    print(f"Document {doc_id}: {text[:100]}...")

Semantic Search Example

import numpy as np
from datasets import load_dataset

# Load dataset
dataset = load_dataset("567-labs/jemail-house-oversight-embeddings")

# Convert embeddings to numpy array
embeddings = np.array([ex["embedding"] for ex in dataset["train"]])

# Query embedding (you would generate this with the same model)
query_embedding = np.array([...])  # Your query embedding

# Compute cosine similarity
similarities = np.dot(embeddings, query_embedding) / (
    np.linalg.norm(embeddings, axis=1) * np.linalg.norm(query_embedding)
)

# Get top-k most similar documents
top_k = 5
top_indices = np.argsort(similarities)[-top_k:][::-1]

for idx in top_indices:
    example = dataset["train"][idx]
    print(f"Score: {similarities[idx]:.4f}")
    print(f"Document: {example['doc_id']}, Page: {example['page_number']}")
    print(f"Text: {example['text'][:200]}...")
    print()

Filtering by Document

from datasets import load_dataset

dataset = load_dataset("567-labs/jemail-house-oversight-embeddings")

# Get all pages for a specific document
doc_id = "EFTA00000001"
doc_pages = dataset["train"].filter(lambda x: x["doc_id"] == doc_id)

print(f"Found {len(doc_pages)} pages for document {doc_id}")

Model Information

• Model: qwen/qwen3-embedding-8b
• Provider: OpenRouter
• Base URL: https://openrouter.ai/api/v1
• Dimension: 8192
• Context Length: 32,768 tokens

Related Datasets

• 567-labs/jemail-doj: Full DOJ documents with metadata
• 567-labs/jemail-house-oversight: Full House Oversight documents with metadata
• 567-labs/jemail-emails: Full email messages with metadata

Citation

If you use this dataset in your research, please cite:

@dataset{jemail_house_oversight_embeddings},
  title={jemail-house-oversight-embeddings},
  author={567-labs},
  year={2025},
  url={https://huggingface.co/datasets/567-labs/jemail-house-oversight-embeddings}
}

License

This dataset is released for research and educational purposes. The underlying documents are public records released by government agencies.

Disclaimer

This dataset contains sensitive content related to criminal investigations. Users should handle this data responsibly and in accordance with applicable laws and ethical guidelines.