Training of the first sentiment model

[roddar92]

Split of sample on train and test:

• shuffle
• 80% of documents - train, 20% - test

Algorithm:

• creating of word2vec
• training model: one of the GBM (i.e. LightGBM) / SVM / ...

Extension of word2vec:

1. Train TF-IDF model (bag-of-words) on the normalized corpus
2. TF = frequency of a word in document, IDF = log(N/#{count of documents with a word})
3. word2vec.get_vector(word) * idf(word)

Document = Tweet