CinPatent: Datasets for Patent Classification

We release two datasets for patent classification in English and Japanese at Google Drive. The data folder structure upon extracted is described in section Structure.

Data description

Each data file is a .ndjson in which each line describes a sample in json format with following attributes.

We partition data with ratio 80:10:10 for training, development, and testing. Following table provides several statistics of our datasets.

Structure

The datasets are available with multiple ratios: 10%, 25%, 50%, 75%, and 100% (in en_patent and ja_patent).

cinpatent
├── en_patent
│   ├── en.ndjson
│   └── segmentation
│       ├── en_0.1.ndjson
│       ├── en_0.25.ndjson
│       ├── en_0.5.ndjson
│       └── en_0.75.ndjson
└── ja_patent
    ├── ja.ndjson
    └── segmentation
        ├── ja_0.1.ndjson
        ├── ja_0.25.ndjson
        ├── ja_0.5.ndjson
        └── ja_0.75.ndjson

Contact

For further support, please contact us at [email protected].