US Granted Patent: Claim 1 Dataset

Dataset Description

This dataset contains the first independent claim from US utility patents spanning 2005-2025. Patent claims define the legal scope of an invention, and the first claim is typically the broadest and most important claim in a patent.

Dataset Structure

Data Fields

• patent_id (string): Unique patent identifier (e.g., "7123456", "D1056398")
• claim_text (string): The text of the first claim, cleaned and normalized

Data Splits

The dataset is organized by year, with each year stored as a separate parquet file (claims_YYYY.parquet).

Data Processing

The dataset was created from PatentsView bulk data with the following processing steps:

1. Filtering: Only claim number 1 (first independent claim) for each granted patent
2. Text cleaning:
   • Removed claim numbering (e.g., "1. ")
   • Normalized whitespace
   • Stripped leading/trailing whitespace
3. Length filtering: Claims between 256 and 3,072 characters (512×6 tokens approximately)

Source Data

• Source: PatentsView Bulk Data
• Coverage: US utility patents granted 2005-2025
• License: PatentsView data is in the public domain

Intended Use

This dataset is intended for:

• Natural language processing research on patent text
• Training language models on technical/legal text
• Analyzing patent claim structure and language evolution
• Prior art search and patent similarity analysis

Limitations

• Only includes the first claim (independent claims), not dependent claims
• Length filtering may exclude very short or very long claims
• Patent IDs include utility patents only (some design/plant patents may be present)
• Text quality depends on USPTO OCR and PatentsView processing