Disambiguation Study for Arabic Applied on Text Classification

Setup

This environment is setup to work on a Linux platform. Make sure to use WSL2 on windows.

• Install developer tools for C++ package building.

sudo apt install build-essential

• Install Java to work with AraBERT preprocessors.

sudo apt install default-jre

• Download and install Miniconda

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -p ./miniconda

• Activate conda base env

source ./miniconda/bin/activate

• Create disambg env from YAML file

conda env create -f environment.yml
conda activate disambg

• Fill .ENV.EXAMPLE file and save it into .env file

• Download Pretrained FastText model for baseline calculations

cd models
python ../download_fasttext_model.py ar
rm cc.ar.300.bin.gz
cd ..

Topic Classification Results

Accuracy

Macro-F1

W&B Workspace

• Workspace Link

• Topic Classification Data Lineage Link