U.S. Patent Claims for Machine Learning Training

Over 1.5 million records of patent claims, likely for training and testing machine learning models, were uploaded by AI-Growth-Lab in July 2022. The dataset appears to contain structured tabular data with text fields, sourced from U.S. patents. It is hosted on Hugging Face with associated tags indicating its scale and intended use for ML tasks.

Use Cases

• Train a text classifier to categorize patent claims by technology area using the claim text field.
• Fine-tune a language model for patent claim summarization or paraphrasing based on the textual content.
• Perform semantic similarity analysis between different patent claims to identify prior art or overlapping inventions.
• Develop a sequence-to-sequence model to generate patent claim drafts from invention descriptions.

Strengths

• Contains over 1.5 million records, providing substantial data for model training.
• Includes text modality, suitable for a range of NLP tasks on legal documents.
• Sourced from U.S. patents, offering a consistent legal and linguistic framework.

Limitations

• Specific column names and data schema are unknown, complicating direct analysis.
• The dataset's creation method and potential preprocessing steps are not documented.
• Last updated in mid-2022, so it may not include recent patents.

Provenance

Source

U.S. patents, inferred from the 'Regionus' platform tag.

Collection Method

Presumably extracted and compiled from patent office databases; exact method is null.

Time Range

null

Freshness

Last updated on 2022-07-31; update frequency is unknown.

Geography

United States, inferred from platform tags.