French Patent Documents From 2024 Parsed Into Paragraphs

35,479 original XML files of French patents published in 2024 were processed and chunked at the paragraph level. The dataset was created by INPI-France, the French intellectual property office, and was last updated in March 2026. It is provided in Parquet format for distributed streaming.

Use Cases

• Train a text classification model to categorize patent paragraphs by technical domain using the chunked text content.
• Fine-tune a language model for patent summarization using the structured <p> tag chunks as input.
• Perform semantic search across the corpus of patent paragraphs to find prior art based on technical descriptions.
• Analyze linguistic patterns and technical terminology evolution within French patent documents from a single year.

Strengths

• Contains text from 35,479 individual patent XML source files.
• Data is pre-chunked at the XML paragraph (<p>) level for immediate processing.

Limitations

• The dataset scope is limited to patents published only in the year 2024.
• Specific details on the total number of text chunks or rows are unavailable.

Provenance

Source

French National Institute of Industrial Property (INPI).

Collection Method

Extracted and structured from original public patent documents (A1, 2024) via API/FTP access.

Time Range

2024

Freshness

Contains data from 2024; dataset page was updated in March 2026.

Geography

France