An entity type enhanced multi-feature fusion model for Chinese mineral exploration named entity recognition

Dataset

We used a Chinese lithium deposit named entity recognition corpus created by ourselves, which includes train.bio, dev.bio, and test.bio files. Each file consists of two columns in the following format:

磷 B-MINE
灰 I-MINE
石 I-MINE
与 O
钠 B-MINE
长 I-MINE
石 I-MINE
、 O
石 B-MINE
榴 I-MINE
石 I-MINE
和 O
白 B-MINE
云 I-MINE
母 I-MINE
共 O
生 O
或 O
伴 O
生 O

Pretrained Model

You need to download the pre-trained model bert-base-chinese，and place

• pytorch_model.bin
• vocab.txt

these two files in ./pretrained_bert_models/bert-base-chinese.

Requirements

• python==3.8.13
• numpy==1.19.3
• pytorch==1.9.1
• pytorch-crf==0.7.2
• tqdm==4.64.1
• transformers==3.1.0

Usage

1. Run encode_labels.py to generate the encoding file for sentences with entity types.
2. Run run.py for training and testing.

References

Partial code implementation was referenced from

• CLUENER
• LEBERT-NER-Chinese