Dataset

DLT-Patents

A collection of 49,023 patent filings focused on distributed ledger technology is available, comprising roughly 1.296 billion tokens. The corpus is a component of the broader DLT-Corpus initiative and is intended to facilitate natural language processing research, innovation analysis, and patent examination within the distributed ledger field.

Open on Hugging Face

Data preview

Subset

Loading preview...

Dataset card

DLT-Patents is a corpus of 49,023 U.S. patent documents related to Distributed Ledger Technology, covering filings from 1990 to 2025 and released by ExponentialScience. It is published in Parquet format with a permissive but unstated license and is hosted on the Hugging Face datasets hub.

What this dataset contains

Each row represents one USPTO patent document drawn from public USPTO XML and full-text databases, deduplicated and normalized to English text. The corpus spans roughly 1.296 billion tokens across the 'text' field and is provided as a single split of 49,025 rows, with users expected to construct their own train, validation, and test partitions. Inventor and assignee names are preserved exactly as recorded by the USPTO, so the dataset reflects official bibliographic information without additional enrichment.

Structure and fields

The schema is entirely string-typed and centers on a long 'text' field holding the extracted patent body. Bibliographic identifiers include 'Patent Number', 'Document ID', 'Family ID', 'Application Number', 'Filing Date', and 'Date Published', while classification is captured through 'CPCI' and 'CPCA' codes. People and parties are tracked via 'Inventor', 'Assignee', 'Applicant Name', 'Primary Examiner', and 'Assistant Examiner', and auxiliary fields 'XREF', 'OR', 'Notes', 'Notes/Tagged', 'Relevancy', 'Database', and 'total_tokens' provide cross-references and per-document metadata.

How analysts use it

The corpus is positioned for NLP research, feature extraction, and innovation studies in the DLT domain rather than as a price or transaction feed. Analysts building crypto or blockchain thematic screens can mine 'Title', 'Assignee', and 'CPCI' codes to map competitive activity, inventor networks, and filing trends tied to specific protocols or applications. Because coverage is U.S.-only and license terms are unstated, users should validate downstream redistribution rights before integrating the data into commercial investment workflows.

Documentation & references 1