PaECTER
About this tool
A patent-document embedding model from the Max Planck Institute for Innovation and Competition. Fine-tuned from BERT for Patents using examiner-added citations, it produces 1,024-dimensional vectors. The paper reports better citation-prediction performance than BERT for Patents and general E5, GTE and BGE models. Uses include patent search, preliminary prior-art screening, clustering and maps.
- Feature 1: Load with sentence-transformers.
- Feature 2: Produce 1,024-dimensional patent embeddings.
- Feature 3: Fine-tuned on examiner citations.
- Feature 4: Companion dataset: mpi-inno-comp/paecter_dataset.
- Feature 5: Colab and Kaggle examples.
Resource information
- Maintainer
- Max Planck Institute for Innovation and Competition; research institute.
- Type
- Dataset / model
- Category
- Patents & intellectual property
- Runs on
- Dataset or model
- Who it's for
- Patent-bibliometrics, innovation-economics and patent-NLP researchers comparing semantic similarity, clustering technologies or screening prior art.
- License at time of listing
- Apache-2.0
- Source
- Hugging Face