USPTO-3M / README.md
MalavP's picture
added references for database construction
a780b6f verified
metadata
dataset_info:
  features:
    - name: cpc_ids
      dtype: string
    - name: id
      dtype: string
    - name: date
      dtype: string
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 3264482506
      num_examples: 3198360
  download_size: 1341114093
  dataset_size: 3264482506
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

This dataset was generated using Google's BigQuery API. The query is adapted from Appendix A from the work by Lee and Hsiang. The query is changed to include patents from 2000 - 2015. The specific query is shown below.

SELECT STRING_AGG(distinct t2.group_id ORDER BY t2.group_id) AS cpc_ids, 
       t1.id, t1.date, t3.text
FROM `patents-public-data.patentsview.patent` t1,
     `patents-public-data.patentsview.cpc_current` t2,
     `patents-public-data.patentsview.claim` t3
WHERE t1.id = t2.patent_id
  AND t1.id = t3.patent_id
  AND timestamp(t1.date) >= timestamp('2000-01-01')
  AND timestamp(t1.date) <= timestamp('2015-12-31')
  AND t3.sequence = '1'
  AND t1.type = 'utility'
GROUP BY t1.id, t1.date, t3.text