ArabicNER

The winning solution to the Topcoder Arabic NER challenge.

Environment setup

nvidia-docker build -t arabic_ner .
nvidia-docker run -v ``pwd''/data:/data:ro -v <Output Path>:/wdata -it arabic_ner

Inference with pre-trained model

bash test.sh /data /wdata/solution.csv

NER Model Training

bash train.sh /data

Inference with trained NER model

Using the following command to use the full model (changing auto to -1 to use CPU):

bash test.sh /data /wdata/solution.csv auto

Or, using the following command to use a single model (changing auto to -1 to use CPU):

bash naive_inference.sh /data /wdata/solution.csv auto

Reference

Liyuan Liu, Jingbo Shang and Jiawei Han. “Arabic Named Entity Recognition: What Works and What’s Next” in Proc.of the 4th Arabic Natural Language Processing Worksho (WANLP 2019), co-located with ACL 2019, Florence, Italy, July2019.