Big Patent: 1.3 Million U.S. Patent Documents and Summaries

BIGPATENT consists of 1.3 million U.S. patent documents paired with human-written abstractive summaries, categorized into nine technical domains. Created by researchers at Northeastern University and published in 2019, it serves as a benchmark for large-scale summarization of technical and legal text.

Use Cases

• Abstractive summarization training using patent descriptions and human-written abstracts
• Topic classification using the nine CPC category codes
• Technical language modeling for legal and engineering domains

Strengths

• 1.3 million record count
• Human-written abstractive summaries
• Nine CPC classification categories for multi-task learning

Limitations

• Geographic bias restricted to U.S. patent filings
• Highly technical and legalistic language profile may not generalize to common prose
• Temporal staleness relative to recent patent filings

Provenance

Source

Northeastern University (Arxiv 1906.03741)

Collection Method

Extracted from official U.S. patent application records

Freshness

Content reflects patent filings up to the 2019 publication date; metadata was updated in 2025.

Geography

United States