chinese-coreference-resolution

Chinese coreference resolution, use chinese-roberta-wwm-ext pretrain model on CLUE WSC2020 dataset

Set up

pip install -r requirements.txt
python classifier.py

Dataset

CLUE BENCHMARK: WSC2020

Example:

{"target": {"span2_index": 35, "span1_index": 6, "span1_text": "洋人", "span2_text": "他们"}, "idx": 1, "label": "true", "text": "有些这样的“洋人”就站在大众之间，如同鹤立鸡群，毫不掩饰自己的优越感。他们排在非凡的甲菜盆后面，虽然人数寥寥无几，但却特别惹眼。"}

Model Structure

Finetune

The result is stored in matric.tsv and average evaluation accuracy is around 80 percent.