Competition

Evaluation Criteria

The submissions are evaluated with precision, recall, and F1 (micro and macro averaged). The evaluation script is available at the Github repository. For more details, please refer to the paper.

Terms and Conditions

The data and reference codes are licensed under the CC BY-SA 4.0 License.

The participant needs to name your result file as ''results.jsonl'' and compress it into zip format for submission.

Each line in the submission file should be a json string encoding the prediction results for one document. The json format is as below:

{

    "id": "6b2e8c050e30872e49c2f46edb4ac044", // id for the document
    "predictions":[ // a list, prediction results for the provided candidates

        {
            "id": "46348f4078ae8460df4916d03573b7de", // id for the candidate
            "type_id": 10, // integer id for the predicted type, 0 for the negative instances
        },

    ]
}