DLT-Patents

Paper | Code

Dataset Description

Dataset Summary

DLT-Patents is a comprehensive corpus of patent documents related to Distributed Ledger Technology (DLT). This dataset is part of the larger DLT-Corpus collection, designed to support NLP research, innovation studies, and patent analysis in the DLT domain.

The dataset contains 49,023 patent documents with 1,296 million tokens (1.296 billion tokens), spanning patents from 1990 to 2025. All documents are in English and sourced from the United States Patent and Trademark Office (USPTO).

This dataset is part of the DLT-Corpus collection. For the complete corpus including scientific literature and social media data, see: https://huggingface.co/collections/ExponentialScience/dlt-corpus-68e44e40d4e7a3bd7a224402

Languages

English (en)

Dataset Structure

Data Fields

Each patent document in the dataset contains the following fields:

• Patent Number: USPTO patent number (e.g., US10123456B2)
• Document ID: Unique document identifier
• Title: Title of the patent
• text: Complete patent text including abstract, claims, and description
• Date Published: Date the patent was published
• Filing Date: Date the patent application was filed
• Family ID: Patent family identifier
• Application Number: Patent application number
• Inventor: List of inventors
• Assignee: Patent assignees (companies or individuals holding the patent)
• Applicant Name: Name of the patent applicant
• Primary Examiner: Name of the primary patent examiner
• Assistant Examiner: Name of the assistant examiner
• CPCI: Cooperative Patent Classification Invention codes
• CPCA: Cooperative Patent Classification Additional codes
• OR: Original reference information
• XREF: Cross-reference information
• Relevancy: Relevancy score or classification
• Notes: Additional notes about the patent
• Notes/Tagged: Tagged notes or annotations
• Database: Source database (USPGPUB or USPAT)
• total_tokens: Total number of tokens in the document

Data Splits

This is a single corpus without predefined splits. Users should create their own train/validation/test splits based on their specific research needs. Consider temporal splits to avoid data leakage in time-series analyses.

Dataset Creation

Curation Rationale

DLT-Patents was created to address the lack of large-scale, domain-specific patent corpora for NLP and innovation research in the Distributed Ledger Technology field. Patents provide unique insights into:

• Commercial applications of DLT technology
• Technical innovations and their evolution
• Industry trends and competitive landscapes
• The transition from research to practical implementation

Source Data

Data Collection

Patents were retrieved from USPTO public databases, specifically:

• USPGPUB: Published patent applications
• USPAT: Granted patents

Data Processing

The collection process involved:

1. Text extraction: Extracting text from USPTO XML and full-text databases
2. Formatting standardization: Normalizing text format and structure
3. Encoding correction: Fixing character encoding errors and special characters
4. Deduplication: Removing duplicate entries and ensuring unique patents
5. Quality filtering: Removing incomplete or corrupted documents

Personal and Sensitive Information

This dataset contains only publicly available patent documents from the USPTO. Inventor and assignee names are retained as they appear in official patent records, which is standard practice for patent documentation. No personal or confidential information beyond what is in public patent records is included.

Considerations for Using the Data

Discussion of Biases

Potential biases include:

• Geographic bias: Only US patents are included; international patents are not represented
• Language bias: Only English-language patents are included
• Temporal bias: More recent years have significantly more patents due to the growth of DLT technology
• Entity bias: Large corporations may be over-represented compared to individual inventors
• Technology bias: Certain DLT applications (e.g., cryptocurrency) may be over-represented compared to others

Other Known Limitations

• USPTO only: Dataset only includes US patents, missing international innovations
• Temporal lag: Recently filed patents may not yet be published (18-month publication delay)
• Keyword limitations: Some relevant patents may be missed due to evolving terminology
• Legal complexity: Patent language is highly technical and legally precise, which may limit general NLP applicability

Additional Information

Dataset Curators

Walter Hernandez Cruz, Peter Devine, Nikhil Vadgama, Paolo Tasca, Jiahua Xu

Licensing Information

Public Domain under USPTO's Terms of Service (TCS).

Patent text is typically not subject to copyright restrictions per USPTO's Terms of Service. Users are free to use, reproduce, and distribute this data. However, users should note:

• The content of patents (the inventions themselves) may be protected by patent rights
• Using patented technologies may require licensing from patent holders
• This dataset provides access to patent text for research purposes, not rights to use patented technologies

For more information, see: https://www.uspto.gov/terms-use-uspto-websites

Citation Information

@misc{hernandez2026dlt-corpus,
      title={DLT-Corpus: A Large-Scale Text Collection for the Distributed Ledger Technology Domain}, 
      author={Walter Hernandez Cruz and Peter Devine and Nikhil Vadgama and Paolo Tasca and Jiahua Xu},
      year={2026},
      eprint={2602.22045},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.22045}, 
}