PatCit: Patent Citations Including Non-Patent Literature and In-Text References

A March 2020 project presentation details the current state. The dataset contains classified Non Patent Literature citations and extracted in-text references from USPTO patents, created by Gaétan De Rassenfosse at École Polytechnique Fédérale de Lausanne. It is designed to uncover interactions in the innovation system beyond front-page citations.

Use Cases

• Study innovation system interactions based on classified Non Patent Literature citations.
• Analyze the relationship between science and patents based on extracted in-text bibliographical references.
• Train NLP models for citation classification based on the 40 million processed NPL citations.
• Examine worldwide patent citation patterns based on the project's stated comprehensive scope.

Strengths

• Classified 40 million Non Patent Literature citations into 9 distinct research-oriented classes with a reported 90% accuracy rate.
• Parsed and consolidated 27 million NPL citations classified as bibliographical references.
• Extracted, parsed, and consolidated in-text references from the body of all-time USPTO patents.
• Uses permissive licenses (MIT-2 for code, CC4 for data) and is designed for community improvement.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

DOCDB database and USPTO patents.

Collection Method

Processed using state-of-the-art machine learning and NLP tools for classification, parsing, and consolidation.

Geography

Worldwide patents, with specific extraction from all-time USPTO patents.