This repository contains research work on finding prior art for a given patent. The approach is to find the most similar documents for a given patent application by comparing them using similarity measures calculated on the documents' full texts. For further details on the experiments please refer to the paper:

@article{helmers2019automating,
  title={Automating the search for a patent's prior art with a full text similarity search},
  author={Helmers, Lea and Horn, Franziska and Biegler, Franziska and Oppermann, Tim and M{\"u}ller, Klaus-Robert},
  journal={{PLoS ONE}},
  volume={14},
  number={3},
  pages={e0212103},
  year={2019},
  publisher={Public Library of Science}
}

All the data sets needed for reproducing the analyses are available at: https://figshare.com and can be downloaded in a compressed format after sign-up

• SQLite database-file: https://figshare.com/articles/Patent_Database/7264733
• Patent scoring by expert and corpus subsample: https://figshare.com/articles/human_eval_tar_gz/7257215
• Entire corpus: https://figshare.com/articles/corpus_tar_gz/7257194

• Adapt the seed patents in the main functions in patentcollector.py

python patentcollector.py

• Save your patent files as .csv files with following metadata as columns: ['id', 'title', 'category', 'pub_number', 'app_number', 'pub_date', 'abstract', 'description', 'claims', 'cited_patents', 'pub_dates']
• Adapt the path in the main function of make_patent_db.py to point to the directory containing your patent files

python make_patent_db.py

Check out the category distributions in your corpus

python compare_cats.py

python idf_regression.py

python kpca.py

python lat_sem_ana.py

python word2vec_app.py

python doc2vec.py

You are free to use the content of this repository under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.