Word Embedding - Hebrew

The Code for https://www.oreilly.com/learning/capturing-semantic-meanings-using-deep-learning

Note: This code works for Hebrew, but it should work on any other language.

1. Download hebrew dataset from wikipedia

   • Go to: https://dumps.wikimedia.org/hewiki/latest/
   • Download hewiki-latest-pages-articles.xml.bz2

   In linux this can be easily done using:

   wget https://dumps.wikimedia.org/hewiki/latest/hewiki-latest-pages-articles.xml.bz2

2. pip install --upgrade gensim (https://radimrehurek.com/gensim/install.html)

3. Run create_corpus.py: python create_corpus.py

   • It will create wiki.he.text

4. train the model: from python prompt:

   • import word2vec
   • word2vec.train()

5. explore model using jupyter notebook. You can use the supplied playingWithHebModel.ipynb example as a starting point.

Word2Vec

• Train (inp = "wiki.he.text", out_model = "wiki.he.word2vec.model")

FastText

pip install fasttext

• Train (inp = "wiki.he.text", out_model = "wiki.he.fasttext.model", alg = "skipgram")

Test

Testing specific Hebrew analogies like:

פריז + גרמניה - צרפת = ברלין

גבר + מלכה - מלך = אישה