Dataset

us_patent_claim1

A collection featuring the primary independent claim of United States utility patents issued between 2005 and 2025 is provided. Because the opening claim usually establishes the widest legal boundaries of an invention, it carries particular significance, and each entry includes the patent identifier alongside its claim text.

Open on Hugging Face

Data preview

Subset

Loading preview...

Dataset card

The US Granted Patent Claim 1 dataset compiles the first independent claim of US utility patents granted between 2005 and 2025. It is published as Parquet files split by year, derived from PatentsView bulk data.

What this dataset contains

The dataset stores the first independent claim text for granted US utility patents across 2005–2025, with one Parquet file per year named `claims_YYYY.parquet`. Filtering retains only claim number 1 and applies text cleaning and length trimming to claims between roughly 256 and 3,072 characters. The single training split lists 4,473,294 rows.

Structure and fields

Two columns are exposed: `patent_id`, stored as a large string identifier for the granted patent, and `claim_text`, also a large string holding the cleaned first-claim language. Because the schema is minimal, downstream users typically join `patent_id` to external patent metadata (assignee, classification, filing date) when richer attributes are needed. Files are provided in Parquet and are compatible with the datasets, Dask, Polars, and mlcroissant libraries.

How analysts use it

Analysts use the corpus for NLP tasks on legal/technical language, such as training or fine-tuning language models on patent claim drafting style. For investment workflows, the joined assignee and classification fields can support tracking inventive activity at specific companies or in particular technology areas, offering a public-records signal of corporate R&D intensity. It can also benchmark text-similarity or classification models on a long historical span of US patent grants.

Documentation & references 1