Datasets:
language:
- en
license: cc-by-sa-4.0
size_categories:
- 100K<n<1M
task_categories:
- text-generation
pretty_name: SEC EDGAR Material Contracts (Exhibit 10)
tags:
- legal
- finance
dataset_info:
features:
- name: index_html_url
dtype: string
- name: index_text_url
dtype: string
- name: cik
dtype: int64
- name: name
dtype: string
- name: type
dtype: string
- name: date
dtype: timestamp[ns]
- name: seq
dtype: int64
- name: desc
dtype: string
- name: doc_type
dtype: string
- name: size
dtype: int64
- name: filename
dtype: string
- name: file_url
dtype: string
- name: file
dtype: string
- name: __index_level_0__
dtype: int64
- name: file_content
dtype: string
splits:
- name: train
num_bytes: 124941581373
num_examples: 930120
download_size: 30356508754
dataset_size: 124941581373
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Material Contracts (Exhibit 10) from SEC/EDGAR
Because sometimes you need 930,120 examples of corporate legalese to train your next model ☕
Table of Contents
- Dataset Card Creation Guide
Dataset Description
- Homepage: NA
- Repository: Scrapy Crawler
- Paper: NA
- Leaderboard: NA
- Point of Contact: mouchenghao at gmail dot com
Dataset Summary
Picture this: 930,120 material contracts (Exhibit 10) painstakingly collected from sec.gov's EDGAR database. We're talking about legal agreements spanning from 1994 to January 2024, sourced from 10-K, 10-Q, and 8-K filings. Think of Exhibit 10 as the treasure trove where companies hide their most important legal paperwork – employment agreements, merger documents, licensing deals, and all those contracts that make corporate lawyers reach for their third espresso.
This isn't just another dataset – it's three decades of corporate America's legal DNA, ready for your next language model to digest.
Supported Tasks and Leaderboards
language-modelingortext-generation: Perfect for building domain-specific models that understand the intricate dance of legal and financial language. Your model will learn to speak fluent "corporate" – complete with all those delightful whereases and heretofores.
Currently, there's no leaderboard for this dataset, but hey, that's an opportunity waiting for someone with enough coffee and determination!
Languages
Primarily US English, though you might encounter the occasional foreign phrase when companies get international. It's like linguistic archaeology – you never know what you'll dig up.
Dataset Structure
Data Instances
Check out the data viewer for examples – trust me, it's more interesting than it sounds. Each instance is like a little corporate story waiting to be told.
Data Fields
Here's what each record contains (because data dictionaries are love letters to future developers):
| Field | Description |
|---|---|
index_html_url |
Filing index page (your breadcrumb trail back to the source) |
index_text_url |
Filing index text page (for when HTML isn't your thing) |
cik |
Central Index Key from EDGAR (think of it as a company's social security number) |
name |
Company name (who's responsible for this legal masterpiece) |
type |
Filing type (10-K, 10-Q, or 8-K – the holy trinity of SEC filings) |
date |
Filing date (when this document saw the light of day) |
seq |
Sequence number in the filing (because order matters) |
desc |
Description provided from the filing (sometimes helpful, sometimes cryptic) |
doc_type |
Document type (e.g. EX-10) (the exhibit classification) |
size |
Document size (bigger isn't always better, but it usually means more billable hours) |
filename |
Document name (often more creative than you'd expect) |
file_url |
Document page URL (your direct line to the source) |
file |
GCS file URI (private, like a good secret) |
__index_level_0__ |
Please ignore (the artifact of data wrangling we all pretend doesn't exist) |
file_content |
Text content (HTML) or base64 string for binary content (PDF) (the good stuff) |
Please be aware, some of the file_content might contain the entire filing as old filings tend to include all exhibits in one txt document.
Data Splits
There's no split by design – everything lives under train because sometimes life is simpler that way. Feel free to create your own splits based on your specific needs (and caffeine levels).
Dataset Creation
Curation Rationale
Why spend sleepless nights building this dataset? Because SEC EDGAR is a goldmine of publicly available corporate intelligence, and someone had to do the heavy lifting of making it ML-ready. This collection of exhibit files gives researchers direct access to the contracts and agreements that drive corporate decision-making – no more parsing through entire filings to find the good stuff.
Source Data
https://www.sec.gov/ (the government's gift to data scientists everywhere)
Initial Data Collection and Normalization
The data collection process was like archaeological excavation, but with more Python and less dirt. We crawled through years of filings (10-K, 8-K, and 10-Q), extracting each exhibit individually with complete metadata. It's the kind of methodical, detail-oriented work that requires multiple monitors and a steady supply of debugging fuel.
Each document was downloaded with surgical precision, preserving all metadata for maximum research utility. Think of it as digital preservation, but for corporate paperwork.
Who are the source language producers?
The unsung heroes behind this dataset are the army of finance professionals, lawyers, and corporate executives who craft these documents. The SEC requires public companies, insiders, and broker-dealers to file these periodic statements, creating a continuous stream of corporate communication that investors and researchers rely on for informed decision-making.
These documents represent the collective voice of corporate America – sometimes eloquent, sometimes bureaucratic, always fascinating from a linguistic perspective.
Annotations
Annotation process
No additional annotations were added – we kept it pure and unadulterated, just the way the SEC intended.
Who are the annotators?
The original authors and their legal teams. Every comma placement was probably billable.
Personal and Sensitive Information
Fair warning: this dataset might contain PII (names, emails, job titles, companies) that's already in the public domain. It comes with the territory when dealing with public filings – executives' names and contact information are part of the transparency requirements.
Considerations for Using the Data
Social Impact of Dataset
This dataset could be the catalyst for better legal and finance language modeling capabilities. Imagine AI systems that can parse complex contracts, identify key terms, or help democratize access to legal understanding. Of course, with great power comes great responsibility – use it wisely.
Discussion of Biases
Given the source and language, this dataset has a distinctly US-centric flavor with a heavy dose of corporate legalese. It reflects the linguistic patterns and legal frameworks of American business culture, which might not translate well to other jurisdictions or informal contexts. Your model might end up sounding like it went to law school (and accumulated the corresponding debt).
Other Known Limitations
Like any dataset scraped from the wild, this one has its quirks. PDF-to-text conversion isn't perfect, OCR sometimes gets creative, and legal documents have their own special way of torturing the English language. Approach with realistic expectations and a good debugger.
Additional Information
Dataset Curators
@chenghao (the sleep-deprived soul who made this possible)
Licensing Information
Attribution-ShareAlike 4.0 International (share and share alike, just the way open source should be)
Citation Information
NA (but feel free to give credit where credit is due)
Contributions
This dataset exists thanks to countless hours of debugging, data wrangling, and the unwavering belief that good datasets make the world a better place. Special thanks to the SEC for maintaining EDGAR and making corporate transparency possible.
Happy modeling! May your training runs be swift and your validation losses ever-decreasing. 🚀
sleepless@debugging:/datasets$ █