Engineering Design Facts from 4,205 USPTO Patents

375,084 example sentences extracted from patents granted by the USPTO, stratified across 130 patent classes. The dataset includes 187,200 positive and 187,884 negative examples, each containing a pair of entities and the engineering design relation between them. It was manually constructed by L. Siddharth at the Singapore University of Technology and Design and last updated on June 16, -2024.

Use Cases

• Train token classification models to identify engineering design entities based on annotated sentence pairs.
• Train Seq2Seq models to generate or populate explicit engineering design relations.
• Analyze the prevalence and structure of design facts across 130 different patent classes.
• Benchmark NLP models for relation extraction tasks within technical and scientific text.

Strengths

• 375,084 manually constructed example sentences provide a substantial text corpus.
• Dataset is balanced with 187,200 positive and 187,884 negative relation examples.
• Source material is a stratified sample of 4,205 patents from the authoritative USPTO.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal and institutional bias inherent to its patent source.

Provenance

Source

Singapore University of Technology and Design (SUTD), via Hugging Face.

Collection Method

Manually constructed from sentences in USPTO patents.

Time Range

Patents granted by USPTO; specific grant years are not provided.

Freshness

Last updated 2024-06-16 01:31:02; freshness should be verified.

Geography

Patent data from the United States Patent and Trademark Office (USPTO).