Our BengWord2Vec is technically a Word2Vec model, which is based on 250 million articles. This is based on https://github.com/Kyubyong/wordvectors, implementation but with additional midifications and updates.

• nltk >= 1.11.1
• regex >= 2016.6.24
• lxml >= 3.3.3
• numpy >= 1.11.2
• gensim > =0.13.1 (for Word2Vec)

• Check this to know how to prepare Word2Vec model for other languages such as Korean, Japanese, French etc.

• STEP 1. Download this raw Bengali texts in case you don't want to collect all the raw text.
• STEP 2. Extract running texts to data/ folder.
• STEP 3. Run build_corpus.py using python3 build_corpus.py
• STEP 4. Run sudo ./make_wordvector.sh to get the BengWord2Vec word vectors but make sure it's already executable.

• Check this for the pre-trained BengWord2Vec model.