COVID-19 Arabic Word Embeddings

We built a word vectors model exploiting our whole COVID-19 dataset collected from January 2020 to April 2020 link. By removing retweets and duplicated tweets, we ended with 2,821,940 tweets. We consider two noticeable word embeddings generation methods: word2vec, and FastText. Using these pre-trained word embeddings models that are domain-specific (COVID-19) would be more accurate than using other generic pre-trained word embeddings in AI tasks.

COVID19 Twitter Word2Vec models

We release four embedding models

Model	Vocabularies No.	Vec-Size	Download
Word2Vec-Twitter-SkipGram	262,715	200	Download URL
Word2Vec-Twitter-CBOW	262,715	200	Dwonload URL
Word2Vec-Twitter-SkipGram	262,715	300	Download URL
Word2Vec-Twitter- CBOW	262,715	300	Dwonload URL

COVID19 Twitter FastText models

We release two embedding models

Model	Vocabularies No.	Vec-Size	Download
FastText-Twitter-SkipGram	262,715	200	Download URL
FastText-Twitter-SkipGram	262,715	300	Download URL

---

Word Embeddings in 2D using T-SNE

Here is the T-SNE visualisation of the word embedding in 2D. It was done using "Embedding Projector" with 3500 iterations and 15 perplexity.

• T-SNE visualisation of Model trianed by Continous Bag of word with a dimension of 300

• T-SNE visualisation of FastText Model trained with a dimension of 300

---

How to use

These models were built using gensim Python library. For loading and using one of the models, you should install the gensim and nltk :

• install gensim >= 3.4 and nltk >= 3.2 using either pip or conda

pip install gensim nltk

conda install gensim nltk

Code sample

---

References

• We built our word embeddings The Word2Vec-Twitter-Skipgram with dimension 200 in our paper COVID-19: What Are Arabic Tweeters Talking About?, to determine the number of topics.

If you are going to use this model, please cite this work using the following bibtext:

@inproceedings{hamoui2020covid,
  title={COVID-19: What Are Arabic Tweeters Talking About?},
  author={Hamoui, Btool and Alashaikh, Abdulaziz and Alanazi, Eisa},
  booktitle={International Conference on Computational Data and Social Networks},
  pages={425--436},
  year={2020},
  organization={Springer}
}

• The rest of word embeddings models were built to be used with classification algorithm in our paper Eating Garlic Prevents COVID-19 Infection: Detecting Misinformation on the Arabic Content of Twitter If you are going to use our pre-trained models, please cite this work using the following bibtext:

@article{alqurashi2021eating,
  title={Eating Garlic Prevents COVID-19 Infection: Detecting Misinformation on the Arabic Content of Twitter},
  author={Alqurashi, Sarah and Hamoui, Btool and Alashaikh, Abdulaziz and Alhindi, Ahmad and Alanazi, Eisa},
  journal={arXiv preprint arXiv:2101.05626},
  year={2021}
}